Huascar Sanchez Posts

Publications

2025-10-25T00:00:00+00:00

(Updated October 2025)

Curriculum Vitae

2017-02-01T00:00:00+00:00

JavaScript is required to view this page.

(Updated October 2025)

Huascar Sanchez

Research Summary and Skills

AI researcher and technical leader at the intersection of AI and software engineering. Proven expertise in building multi-LLM systems, compound AI workflows, and large-scale systems for Big Code analytics, programming language processing, and program analysis (static and dynamic). Current research interests center on multi-LLM collaboration optimization and LLM chemistry estimation as methods for improving specification learning, code generation, and program repair — with a focus on cost-optimized, workflow-aware planning for compound software tasks. I apply AI assurance and AI red teaming to ensure these systems are reliable and robust in high-stakes domains. Strong track record of translating AI/ML research into practical systems, including rigorous automated evaluation frameworks for model assessment. Additional background in modeling, analyzing, and detecting social-cyber operations against open-source ecosystems using deep learning.

Skills

Multi-LLM Collaboration Optimization

Model Evaluation & Benchmarking

Compound AI & Workflow-aware planning

Deep Learning Model Development & Training

LLM Chemistry Estimation for Model Strategy & Inference Optimization

Data Mining (Programming Language Processing)

Automated Software Engineering

LLM workflows for Coding Automation

AI Assurance & AI Red Teaming

AI/ML on Source Code

Static & Dynamic Program Analysis

Social-Cyber Security

Research Interests

Multi-LLM collaboration optimization and LLM chemistry estimation.

Social-cyber security for open-source ecosystem integrity.

AI assurance and error profiling for unexpected model failures.

Static/dynamic program analysis for code analysis/transformation/AI-driven reasoning.

Projects

2025 – now

Reliable Multi-AI Collaboration (Team: SRI)
Leading the development of a post-training framework for optimizing model performance for complex software tasks via LLM Chemistry estimation, optimal multi-LLM collaboration, tool integration via MCP, prompt strategies, and probabilistic multi-LLM consensus mechanisms for uncertainty quantification to ensure the reliable, consistent, and correct execution of complex reasoning tasks.

2024 – now

ARPA-H PARADIGM (Team: SRI)
Developing multi-LLM strategies and workflows for extracting data types and constraints, inferring JSON schemas, and synthesizing schema-compliant code from diverse healthcare data formats to enable automatic cross-device-data-format translation. Includes profiling LLM failure modes to diagnose root causes for hallucinated constraints and brittle formatting.

2024 – now

IARPA ReSCIND (Team: SRI, IHMC, TwoSix, UFL, RAD, GMU, ColumbiaU, SimSpace)
Leading the development of Cyber-Specific Computational Cognitive Models using Self-Supervised Machine Learning, Information Foraging Theory, and Fuzzy Cognitive Maps.

2024 – now

ARPA-H DIGIHEALS (Team: SRI, Confidencial Inc., Homeward Health)
Advancing the automatic generation of formal representations from natural language using neuro-symbolic AI and retrieval augmented generation (RAG), and leading the development of multi-LLM (Agentic) workflows for healthcare data analysis, formalization, and knowledge composition.

Previous projects^*

* Order of projects is arbitrary.

DARPA ARCOS, DARPA SocialCyber, DARPA SoSITE, DARPA MUSE, DARPA CSFV.

Education

2015

University of California Santa Cruz · Ph.D. in Computer Science
Dissertation: Source code Curation Tooling for the Code Forager
Advisor: Prof. Jim Whitehead

2006

San Jose State University · M.S. in Software Engineering
MS Thesis title: Building Systems Using Patterns: Creating Knowledge Maps
Advisor: Prof. M.E. Fayad

2001

Catholic University of Nicaragua · B.S. in Information Systems Engineering

Professional Employment

2024 – 2025

SRI International · Menlo Park, CA
Senior Computer Scientist, Computer Science Laboratory
Spearheading research into inference-time optimization for compound software tasks such as automated program repair and code generation, developing methodologies to improve model reliability, consistency, and performance, focusing on optimal multi-LLM collaboration, multi-hop reasoning, tool integration via MCP, probabilistic multi-LLM consensus, LLM chemistry estimation, prompt optimization, resulting in higher accuracy and reduced hallucinations compared to single-model inference.

2021 – 2023

SRI International · Menlo Park, CA
Advanced Computer Scientist II, Computer Science Laboratory
Continuing to lead DARPA SocialCyber research to secure open-source software supply chains through a novel neurosymbolic AI method that models social-cyber threats. Developed multimodal, multi-task deep learning models for predicting developer behavior shaping the technical direction of open-source software. Recruited, hired, and supervised research intern, providing daily technical mentorship and conducting formal performance evaluations.

2019 – 2021

SRI International · Menlo Park, CA
Advanced Computer Scientist I, Computer Science Laboratory
Guiding the integration of formal verification and program analysis for continuous assurance case construction. Leading DARPA SocialCyber as Principal Investigator, shaping project vision, delivering hybrid AI methods for securing software supply chains, mentoring junior researchers, and coordinating reporting across stakeholders.

2018 – 2019

SRI International · Menlo Park, CA
Advanced Computer Scientist, Computer Science Laboratory
Applying machine learning, large-scale graph mining, and dynamic program analysis to open-source software security, and developing integration and verification methods for systems of systems in distributed environments.

2017 – 2018

SRI International · Menlo Park, CA
Computer Scientist, Computer Science Laboratory
Leveraging \textit{Big Code} analytics and machine learning on source code for code-to-code search, API recommendation, code synthesis, and source code analysis. Managed and coached research intern on graph-based code analysis project, overseeing technical work.

2015 – 2017

SRI International · Menlo Park, CA
Postdoctoral Fellow, Computer Science Laboratory, Host: Hassen Saïdi
Developing learning-to-rank models for large-scale code search and recommendation, and lightweight source code summarization techniques for improved source code understanding.

2014 – 2015

SRI International · Menlo Park, CA
Visiting Research Fellow, Computer Science Laboratory, Host: John Murray
Making formal verification scalable and accessible to broader audiences via crowd-sourced approaches—crowd-sourced formal verification—that enable participation without formal methods expertise.

2013 – 2015

University of California Santa Cruz · Santa Cruz, CA
Research Assistant, Software Introspection Laboratory
Empirical validation of crowd-sourced formal verification methods enabled by online games.

2012 – 2013

University of California Santa Cruz · Santa Cruz, CA
Research Assistant, Information Retrieval Laboratory
Developing latent representation learning methods from social media data for multimodal cyberbullying detection.

2011 – 2012

University of California Santa Cruz · Santa Cruz, CA
Research Assistant, Software Introspection Laboratory
Developing AI-guided, mixed-initiative content creation tools for automatic robot components design.

2010 – 2011

University of California Santa Cruz · Santa Cruz, CA
Research Assistant, Database Group
Advancing semi-automatic index tuning in DBMSs to make it easier and more effective.

2010 – 2012

Infomotor, Inc. · San Francisco, CA
Software Engineer (Consultant)
Streamlining both reporting and presentation of key performance metrics to deliver data that is both understandable and actionable.

2006 – 2009

Pearson VUE · Bloomington, MN
Software Engineer
Developing and maintaining distributed applications that deliver over four million computer-based tests annually worldwide, supporting clients in licensure, certification, and assessment markets.

Publications

Conference Proceedings

2024

2016

JayHorn: A Framework for Verifying Java programs.
T. Kahsai, P. Rümmer, H. Sanchez, and M. Schäf
CAV: 28th International Conference on Computer Aided Verification, 2016.

2016

Detecting Similar Programs via the Weisfeiler-Lehman Graph Kernel.
W. Li, H. Saïdi, H. Sanchez, M. Schäf, and P. Schweitzer.
ICSR: 15th International Conference on Software Reuse, 2016.

2016

Multistaging to Understand: Distilling the Essence of Java Code Examples.
H. Sanchez, J. Whitehead, and M. Schäf.
ICPC: 24th IEEE International Conference on Program Comprehension, 2016.

2015

Source Code Curation on StackOverflow: The Vesperin System.
H. Sanchez and J. Whitehead.
ICSE: 37th IEEE/ACM International Conference on Software Engineering, 2015.

2014

2013

SNIPR: Complementing Code Search with Code Retargeting Capabilities.
H. Sanchez.
ICSE: 35th International Conference on Software Engineering, 2013.

2012

Kaizen: A Semi-Automatic Index Advisor.
I. Jimenez, H. Sanchez, Q.T. Tran, and N. Polyzotis.
SIGMOD: The 2012 ACM International Conference on Management of Data, 2012.

2012

Twitter Bullying Detection.
H. Sanchez and S. Kumar.
NSDI: Proceedings of the 8th USENIX Conference on Networked Systems Design and Implementation, 2012.

2010

2006

The Branding Analysis Pattern.
M. E. Fayad, H.A. Sanchez, and S.K. Singh.
AICCSA: Proceedings of the IEEE International Conference on Computer Systems and Applications, 2006.

2005

Towards Scalable and Adaptable Software Architectures.
M.E. Fayad, H.S. Hamza, and H.A. Sanchez.
IRI: The 2005 IEEE International Conference on Information Reuse and Integration, 2005.

2005

The Dynamic Mapping Design Pattern.
J. Gresh, J. McKim, and H. Sanchez.
PLOP: 12th Pattern Language of Programs Conference, 2005.

2004

A Pattern Language for CRC Cards.
M. Fayad, H. Sanchez, and H. Hamza.
PLOP: 11th Conference on Pattern Languages of Programs, 2004.

2003

The Sampling Analysis Pattern.
H.A. Sanchez, B. Lai, and M.E. Fayad.
IRI: IEEE International Conference on Information Reuse and Integration, 2003.

Workshop Papers

2023

Software Introspection for Signaling Social-Cyber Operations.
H. Sanchez and B. Hitaj.
DESTION: 5th Workshop on Design Automation for CPS and IoT, 2023.

2016

Putting Your Passwords on Self Destruct Mode: Beating Password Fatigue.
H. Sanchez, J. Murray, and D. Sanchez.
SOUPS: 12th Symposium on Usable Privacy and Security, 2016.

2015

Journal Articles

2011

Benchmarking Online Index-Tuning Algorithms.
I. Jimenez, J. LeFevre, N. Polyzotis, H. Sanchez, and K. Schnaitter.
IEEE Data Engineering Bulletin, 2011.

Books

2014

Software Patterns, Knowledge Maps, and Domain Analysis.
M. Fayad, H.A. Sanchez, S.G.K. Hegde, A. Basia, and A. Vakil.
CRC Press, 2014. ISBN 9781466571433

Reports and Other Publications

2025

Task Explorer Pipeline (TEP): Revealing the Hidden Structure of Human Problem-Solving.
T. Odem, G. Denker, and H. Sanchez.
(Under Review)

2025

Multi-LLM Collaboration for Medication Recommendation.
H. Sanchez, B. Hitaj, J. Bergmann, and L. Briesemeister.
ArXiv [Cs.LG]. arXiv:2512.05066 (Under Review)

2025

LLM Chemistry Estimation for Multi-LLM Recommendation.
H. Sanchez and B. Hitaj.
ArXiv [Cs.LG]. arXiv:2510.03930 (Under Review)

2022

Trust in Motion: Capturing Trust Ascendancy in Open-Source Projects using Hybrid AI.
H. Sanchez and B. Hitaj.
ArXiv [Cs.SE]. arxiv:2210.02656

2021

Leveraging Team Dynamics to Predict Open-source Software Projects' Susceptibility to Social Engineering Attacks.
L. Giovanini, D. Oliveira, H. Sanchez, D. Shands.
ArXiv [Cs.SE]. arxiv:2106.16067

Patents

2023

Intelligent Collaborative Project Management.
D. Sanchez, H. Sanchez, H. Saïdi.
United States patent 11610173, 2023.

2022

Multipath verification of data transforms in a system of systems.
B. Dutertre, S. Jha, H. Sanchez, P. Lincoln, E. M. Pearson, R. Dean, I. A. Mason.
United States patent 2022/0197881, 2022.

Talks

2025

Collaborative AI for Data Type Extraction and JSON Schema Learning.
ARPA-H PARADIGM, Virtual meeting, June, 2025.

2025

Multi-level Cyber-Specific Computational Cognitive Models (C3Ms).
IARPA ReSCIND, Virtual meeting, May, 2025.

2023

Software Introspection for Signaling Social-Cyber Operations.
5th Workshop on Design Automation for CPS and IoT, TX. 2023.

2022

Graph-based Pattern Mining in Code Revisions for API Misuse Detection.
DARPA SocialCyber, Virtual meeting, 2022.

2022

Temporal Developer Activity Embeddings for Modeling Evolving Developer Behavior in Social Coding Environments.
DARPA SocialCyber, Virtual meeting, 2022.

2022

SIGNAL: Protecting the Integrity of Open-source Projects using Hybrid AI.
Computer Science Laboratory's Future Perfect Seminar, Virtual meeting. SRI International, Menlo Park, CA, 2022.

2021

A Multi-Task Learning Approach for Software Development Tasks Prediction.
DARPA SocialCyber, Virtual meeting, 2021.

2021

Change Point Analysis of Evolutionary Signatures in Social-Cyber Attacks.
DARPA SocialCyber, Virtual meeting, 2021.

2020

Continuous Certification of Open-source Software using Baseline DesCert.
DARPA ARCOS, Virtual meeting, 2020.

2019

Applying Machine Learning to Open-Source Code Retrieval and Verification.
Digital Foundations and Mathematics. Sandia National Laboratories. Livermore, CA, 2019.

2018

API Workflow Recommendations for Codebase Exploration.
DARPA MUSE. University of Texas at Austin, TX, 2018.

2017

Scaling Inter-Project Software Analysis.
DARPA MUSE, Massachusetts Institute of Technology (MIT), MA, 2017.

2015

Source Code Curation Tooling for the Code Forager.
Computer Science Laboratory Seminars. SRI International, Menlo Park, CA, 2015.

Service

Service to the Field

2025

IEEE Secure Development Conference Program Committee.
IEEE SecDev 2025. Call for Papers.

2023

NSF GRFP Reviewer.
NSF Graduate Research Fellowship Program (GRFP 2023). NSF 23-605

2022

Mining Challenge Program Committee.
19th International Conference on Mining Software Repositories (MSR 2022), Pittsburgh, PA.

2020

Artifact Evaluation Program Committee.
International Symposium on Software Testing and Analysis (ISSTA 2020), Los Angeles, CA.

2017

Paper Reviewer.
The 9th NASA Formal Methods Symposium (NFM 2017), Moffett Field, CA.

2016

Paper Reviewer.
The 13th International Colloquium on Theoretical Aspects of Computing (ICTAC 2016), Taipei, Taiwan.

2014

Paper Reviewer.
The 11th Working Conference on Mining Software Repositories (MSR 2014), Hyderabad India.

Mentorship

2022

Akshith Gunasekaran. PhD Student (Oregon State University)
Summer Internship, SRI International, Menlo Park, CA.

2017

Di Yang. PhD Student (University of California Irvine)
Summer Internship, SRI International, Menlo Park, CA.

2015

Xu Zhang. PhD Student (University of California Santa Cruz).

Software Development^*

* Some projects listed above are private on GitHub; please contact me for access details.

Committer

Lapis
Automated Red Teaming framework to evaluate AI robustness against emergent behaviors by simulating persuasion engineering, deception, prompt abuse, and sandbagging attacks;https://github.com/hsanchez/lapis (Python, PyTorch, Pydantic AI, Hugging Face (Transformer), OpenAI and Anthropic SDK).

LLM Chemistry
LLM Chemistry Estimation for Multi-LLM Recommendation; https://github.com/SRI-CSL/llm-chemistry (Python, PyTorch, Pandas, Beartype, SciPy).

Aligner
LLM Workflows for JSON Schema learning; https://github.com/SRI-CSL/aligner (Python, PyTorch, Langchain, HuggingFace (Transformer), Docker, Anthropic's MCP SDK).

Collaborative LLMs
Optimal Multi-AI Collaboration; https://github.com/hsanchez/SRI-CSL/collm (Python, PyTorch, Ollama, LangChain, LangGraph).

Yellow-Patch
A source code and patch analysis library; https://github.com/hsanchez/yellow-patch (Python, PyTorch, HuggingFace (Accelerate), Pandas, SQL, Docker).

Evidential Tool Bus
A framework for scaling assurance case construction, program verification and analysis; https://github.com/SRI-CSL/etb2 (Java, Python).

DesCert
A framework for continuous safety & security evidence generation, curation and assurance case construction; https://github.com/SRI-CSL/DesCert (Java, Python, C).

SIGNAL
Securing open-source software supply chains against social-cyber threats using hybrid AI methods; https://github.com/SRI-CSL/signal-public (Python, PyTorch, Pandas, HuggingFace (Transformer, Accelerate), Postgres, Docker).

SENSOR
Graph-based pattern mining in code revisions for API misuse detection; https://github.com/SRI-CSL/signal-sensor (Python, LLVM, Hydra).

SimProg
Inter-project program similarity detection using graph kernels; https://github.com/SRI-CSL/simprog (Python, PyTorch).

JayHorn
Model checking for Java programs; https://github.com/jayhorn (Java).

do-like-javac
Program analysis and verification pipeline for Java projects; https://github.com/SRI-CSL/do-like-javac (Java, Scala, C, Python).

Callspace
Reliable API recommendation using likely program invariants and graph kernels; https://github.com/SRI-CSL/smac/callspace (Scala, Java).

Randoop Gradle Plugin
Gradle plugin for running Randoop unit test generator; https://github.com/SRI-CSL/randoop-gradle-plugin (Kotlin, Java).

Daikon Gradle Plugin
Gradle plugin for running Daikon invariant detector; https://github.com/SRI-CSL/daikon-gradle-plugin (Kotlin, Java).

Contributor

CHEKOFV
A framework for crowd-sourced formal verification; https://github.com/SoftwareIntrospectionLab/chekofv (Python, Nodejs, TypeScript, ActionScript).

PASCALI
High-volume structured and unstructured data analysis for program analysis, code search, and program synthesis; https://github.com/SRI-CSL/pascali (Python, Scala, Java, SQL).

Read to Understand (RTU) Task

2016-02-22T00:00:00+00:00

Please follow all of the instructions below:

PREREQS

Make sure you have

Google Chrome latest version installed on your computer.
Uninstalled any previously installed version of Violette from Chrome.

GUIDELINES FOR THE READ TO UNDERSTAND

GOAL

Gain understanding in (without multi staging to understand)

Control flow: what is the sequence of execution?
State: what is the content of a data object?
Operations: what does the code compute?
Data flow: where does a data object get updated?
Function: what is the overall functionality?

GUIDELINES

Inspect each code example (one at a time).
Spend your time wisely, don’t go outside the code example’s Q&A page.
Gain understanding in the asked abstractions; e.g., control flow, etc.

THE TASK

There will be 3 Java code examples (Accepted Answers) to review and understand. These examples were extracted from StackOverflow. Your task is to use the Read to Understand (RTU) comprehension technique to dig into each of the Java code examples. This technique involves reading the source code of the given code examples. After 30 minutes, per Java code example, you will be asked to answer a number of questions about the reviewed piece of code.

Please record the time spent reviewing the Java code example. 
Please submit this information as the answer for "How long did you 
spend reviewing and understanding the Java code example?", 
which is found in the online questionnaire you will be 
completing after finishing the whole task.

For each example, there will be two links. First link will take you to a page containing a StackOverflow’s accepted answer. This answer contains some Java code. Second link will take you to an online questionnaire where you’ll answer a number of program comprehension questions about the reviewed piece of code.

After you’ve finished reviewing and understanding the Java code example, please comeback to the experiment page. Then, click on the Java code example’s second link to answer the program comprehension questionnaire.

After answering the questionnaire, close its page and then comeback to the experiment page. Once there, click on the next Java code example to review.

Lastly, after reviewing all code examples and answering their questionnaires, please answer our debriefing questionnaire.

TASK EXPECTATIONS

We expect the following throughout the task:

1. We expect you to review and understand a total of 3 Java code examples, in the order provided in this page.

2. We expect the whole task be done within 2 hrs; including the the time spent reviewing this page.

3. We expect the subjects to use only the resources available on the question-answer page to complete the task; i.e., no external resources like Wikipedia should be used.

LET’S GET STARTED

So without further ado, please click on the first link of Code Example 1 below to start the task:

1. Code Example 1 (code, program comprehension questionnaire)
2. Code Example 2 (code, program comprehension questionnaire)
3. Code Example 3 (code, program comprehension questionnaire)

Multistaging to Understand (MTU) Task

2015-12-05T00:00:00+00:00

Please follow all of the instructions below:

PREREQS

Make sure you have

Google Chrome latest version installed on your computer.
Uninstalled any previously installed Violette from Chrome.
Installed Violette-multi-staging-mode on Chrome¹.

GUIDELINES FOR THE MULTI STAGING TO UNDERSTAND

GOAL

Gain understanding in

Control flow: what is the sequence of execution?
State: what is the content of a data object?
Operations: what does the code compute?
Data flow: where does a data object get updated?
Function: what is the overall functionality?

GUIDELINES

Inspect each code example (one at a time).
Spend your time wisely, don’t go outside the Q&A page.
Click on the stage button² to generate an ordered set of stages³; e.g., 0:stage-name, 1:stage-name, …, N:stage-name
Review the name of the stages in the order they appeared.
Iteratively click on any of the generated code stages to introduce a subset of the given Java code example’s behavior. E.g., Click stages in any order.
4.1 Open any hidden code blocks as you go along, and then try to understand its enclosed elements.
Review the introduced subset of behavior to make sure you understand what it is actually doing.

THE TASK

There will be 3 Java code examples (Accepted Answers) to review and understand. These examples were extracted from StackOverflow. Your task is to use the Multistaging to Understand (MTU) comprehension technique to dig into each of the Java code examples. This technique involves a few steps. Given a series of code stages (extracted from a code example), you will non-sequentially inspect a few of these stages, mentally abstract their functionality, and then use that knowledge to understand the example’s intended function. After 30 minutes, per Java code example, you will be asked to answer a number of questions about the reviewed piece of code.

After you’ve finished reviewing and understanding the Java code example, please click on the Bring it to Desktop button and then provide a brief description of the (1) goal behind this code example and (2) the reason for using it. Then, please press the _enter_key to record your answer.

Please comeback to the experiment page after bringing the Java code example onto your desktop. Then, click on the Java code example’s second link to answer the program comprehension questionnaire.

After answering the program comprehension questionnaire, close its page and then comeback to the experiment page. Once there, click on the next Java code example to review.

Lastly, after reviewing all code examples and answering their questionnaires, please answer our debriefing questionnaire.

TASK EXPECTATIONS

We expect the following throughout the task:

1. We expect you to review and understand a total of 3 Java code examples, in the order provided in this page.

2. We expect the whole task be done within 2 hrs; including the the time spent reviewing this page.

3. We expect the subjects to use only the resources available on the question-answer page to complete the task; i.e., no external resources like Wikipedia should be used.

LET’S GET STARTED

Before starting, please watch the following video. This video introduces violette and the multi-staging functionality and UI.

So without further ado, please click on the first link of Code Example 1 below to start the task:

1. Code Example 1 (code, program comprehension questionnaire)
2. Code Example 2 (code, program comprehension questionnaire)
3. Code Example 3 (code, program comprehension questionnaire)

Open Google Chrome and on the Url bar type chrome://extensions. Look for violette-staging-no-summarization.crx. Get the file and then drop it on the Chrome extension area. Press the Add button and you’re done. ↩
This button is located on the top right corner of Violette. Each stage is represented as named button. Stages are presented in ascending order; i.e., starting from 0 to n where n is the maximum number of stages that can be extracted from the code example. ↩
If after clicking on the stage button you get an error (red box) from the server, then please refresh the Web page and retry to stage the example. ↩

Data

2015-12-05T00:00:00+00:00

Data set

We collected 50,000 Q\&A pages containing Java code examples, accumulated over a 4 year span. Our data set is available at stackoverflow-corpus.

We sorted this data set into 3 size categories: short (~50 LOC), medium (~135 LOC), and long (~200 LOC). Then, we randomly selected the examples from each size category.

MTU/RTU answers

The answers to the open ended program comprehension questions are available at program-comprehension-answers.

Source Code Curation Tooling for the Code Forager

2015-10-25T00:00:00+00:00

Abstract

The Web has changed the dynamics of programming. We are in an era where reusing code from the Web is frequently the norm, not the exception. In this era, programmers use question and answer (Q&A) sites like StackOverflow to pose coding questions and receive coding answers. Programmers using these sites often engage in a complex code foraging process of understanding and adapting the code snippets they encounter to determine their fitness for use. While search still dominates modern code retrieval, search alone offers little support for validating search results for fitness of use. This is, in large part, due to the inherently questionable quality of online source code. Most online source code is not guaranteed to be good, to work properly, or to be trustworthy. This dissertation focuses on this challenge by introducing Source Code Curation, along with a set of tools that implements it. Source Code Curation is a blend of filtering, refinement, and validation activities. Source Code Curation can help programmers determine what source code is more likely to be useful, and what’s not. Specifically, it can help them both address the inherently questionable quality of online source code upfront, and complete their source code understanding tasks quickly and accurately.

Codepacking

2015-04-04T00:00:00+00:00

Codepacking

Codepacking is a technique for automatically resolving incomplete code snippets. Its ultimate goal is not about achieving program correctness but approximating program completeness. Codepacking produces an Abstract Syntax Tree (AST) of a partial program that Vesperin can manipulate.

Codepacking consists of four steps:

Surrounds code example with the appropriate body declarations (class, method, or both), if needed.
It cross references missing type information of existing ill-typed expressions with a repository of pre-acquired API information and then suggests their well-typed corrections.
It uses Vesperin’s code transformation utilities to add the suggested corrections to the example’s code.
It produces the desired AST.

Codepacking’s code is still private. However, once it becomes public, it will be available here.

My book is out!

2015-03-05T00:00:00+00:00

Software Patterns, Knowledge Maps, and Domain Analysis

My book is about domain analysis and software patterns. Briefly, it shows how to build software systems as systems of patterns, called Knowledge Maps.

It is now in the global markets. Its publication date was December 2014. Please check it out:

Figure 1. Front cover of my book

The citation (in BibTex) of this book is presented as follows:

@book{hsanchez, 
      author = {M. E. Fayad, H. A. Sanchez, S. G .K. Hegde, 
                A. Basia, and A. Vakil.},
      title  = {Software Patterns, Knowledge Maps, and 
                Domain Analysis},
      publisher = {Auerbach Publications},
      year      = 2014,
      address   = {Boca Raton, Florida},
      isbn      = {978-1466571433}
}

Violette with Multi Staging and Code Summarization

2015-03-04T00:00:00+00:00

To evaluate Violette’s functionality, I have put together the following experiment. Please follow all of the instructions below:

PREREQS

Make sure you have

Google Chrome latest version installed on your computer.
Uninstalled any previously installed Violette from Chrome.
Installed Violette-staging-mode on Chrome¹.

THE TASK

You, the subject, must click on a linked code example below. This link will take you to a page containing a StackOverflow’s accepted answer. This answer contains some Java code.
Once you are on that page, your goal is to review and understand the Java code example.

If the links takes you to an answer without any Java code, feel free to comeback to experiment page and then refresh it. This action will give you a new linked code example.

For this task, you will be using both Violette’s multi-staging functionality² and code summarization via automated code folding.

The multi staging functionality is activated by clicking on the stage button. Each stage is represented as named button.

Code summarization functionality is activated when clicking on any of the produced stages. This functionality will identify the location of a few code blocks in the stage and then it will automatically fold them.

Please record on a piece of paper those examples that did not have any Java code inside. Please submit this information as part of a survey you will completing after finishing the experiment.
After you’ve finished reviewing and understanding the code example, please click on the Bring it to Destop button and the provide a description of the goal behind this code example and the reason for using it. Including, fitness of purpose, abstractions, usage protocols, etc.

If you’ve previously received a key or username from me, then include it in the description of the code example; e.g., username: the code does ….
Lastly, bringing the code example to your desktop, please comeback this page and then complete the following survey.

TASK EXPECTATIONS

We expect the following throughout the task:

1. We expect you to review and understand a total of 10 Java code examples generated by this page.

2. We expect the whole task be done within 1 hr and 30 minutes.

3. We expect the subjects to use only the resources available on the question-answer page to complete the task; i.e., no external resources like Wikipedia.

LET’S GET STARTED

So without further ado, please click on the link below to start the task:

Thank you for participating in this experiment. Now, we need your feedback. Please take a few minutes to complete our survey.

Open Google Chrome and on the Url bar type chrome://extensions. Look for vesper-web.crx. Get the file and then drop it on the Chrome extension area. Press the Add button and you’re done. ↩
Multi staged code examples are divided into a series of intermediate phases, presenting lower-level modules (supporting modules) before higher-level modules (main modules). ↩

Multistager

2015-02-15T00:00:00+00:00

Code Multistager, Multistaging to Understand

The idea behind Multistaging to Understand (MTU) is that given a series of code stages (extracted from a code example), programmers non-sequentially inspect a few of these stages, mentally abstract their functionality, and then use that knowledge to understand the example’s intended function. The following slides walks the reader through what MTU is and what’s behind it:

The code of our Code examples multistager is still private. However, once it becomes public, it will be available here.

If you want to get access to this code, please send me an email.

Multi Staging to Understand Task

2015-01-28T00:00:00+00:00

Please follow all of the instructions below:

PREREQS

Make sure you have

Google Chrome latest version installed on your computer.
Uninstalled any previously installed Violette from Chrome.
Installed Violette-multi-staging-mode on Chrome¹.

GUIDELINES FOR THE MULTI STAGING TO UNDERSTAND

GOAL

Gain understanding in

Control flow: what is the sequence of execution?
State: what is the content of a data object?
Operations: what does the code compute?
Data flow: where does a data object get updated?
Function: what is the overall functionality?

GUIDELINES FOR MULTI STAGING²

Spend your time wisely, don’t go outside the Q&A page.
Click on the stage button³ to generate an ordered set of stages⁴; e.g., 0:stage-name, 1:stage-name, …, N:stage-name
Review the name of the stages in the order they appeared.
Iteratively click on any of the generated code stages to introduce a subset of the given Java code example’s behavior. E.g., Click stages in any order.
4.1 Open any hidden code blocks as you go along, and then try to understand its enclosed elements.
Review the introduced subset of behavior to make sure you understand what it is actually doing.

THE TASK

There will be 3 Java code examples (Accepted Answers) to review and understand. Use the appointed comprehension technique (either Read to Understand or Multi staging to Understand) to dig into each of the Java code examples. After 30 minutes, per Java code example, you will be asked to answer a number of questions about the reviewed piece of code.

After you’ve finished reviewing and understanding the Java code example, please click on the Bring it to Desktop button and then provide a brief description of the goal behind this code example and the reason for using it. Then press the enter key.

After bringing the Java code example to your desktop, please comeback to the experiment page. Then, click on the Java code example’s second link to answer the questionnaire.

After answering the questionnaire, close its page and then comeback to the experiment page. Once there, click on the next Java code example to review.

Lastly, after reviewing all code examples and answering their questionnaires, please answer our debriefing questionnaire.

TASK EXPECTATIONS

We expect the following throughout the task:

1. We expect you to review and understand a total of 3 Java code examples, in the order provided in this page.

2. We expect the whole task be done within 2 hrs; including the the time spent reviewing this page.

3. We expect the subjects to use only the resources available on the question-answer page to complete the task; i.e., no external resources like Wikipedia should be used.

LET’S GET STARTED

Before starting, please watch the following video. This video introduces violette and the multi-staging functionality and UI.

So without further ado, please click on the first link of Code Example 1 below to start the task:

1. Code Example 1 (src, q&a)
2. Code Example 2 (src, q&a)
3. Code Example 3 (src, q&a)

Open Google Chrome and on the Url bar type chrome://extensions. Look for violette-staging-no-summarization.crx. Get the file and then drop it on the Chrome extension area. Press the Add button and you’re done. ↩
Multi staged code examples are divided into an ordered set of code stages, such that each code stage builds upon, and in relation to, preceding code stages. Each code stage is a code fragment that captures a subset of the given Java code example’s behavior. ↩
This button is located on the top right corner of Violette. Each stage is represented as named button. Stages are presented in ascending order; i.e., starting from 0 to n where n is the maximum number of stages that can be extracted from the code example. ↩
If after clicking on the stage button you get an error (red box) from the server, then please refresh the Web page and re-try to stage the example. ↩

Read to Understand Task

2015-01-27T00:00:00+00:00

Please follow all of the instructions below:

PREREQS

Make sure you have

Google Chrome latest version installed on your computer.
Uninstalled any previously installed version of Violette from Chrome.

GUIDELINES FOR THE READ TO UNDERSTAND

GOAL

Gain understanding in (without multi staging to understand)

Control flow: what is the sequence of execution?
State: what is the content of a data object?
Operations: what does the code compute?
Data flow: where does a data object get updated?
Function: what is the overall functionality?

GUIDELINES

Spend your time wisely, don’t go outside the Q&A page.
Gain understanding in the asked abstractions; e.g., control flow, etc.

THE TASK

Please record the time spent reviewing the Java code example. 
Please submit this information as the answer for "How long did you 
spend reviewing and understanding the Java code example?", 
which is found in the online questionnaire you will be 
completing after finishing the whole task.

After you’ve finished reviewing and understanding the Java code example, please comeback to the experiment page. Then, click on the Java code example’s second link to answer the questionnaire.

After answering the questionnaire, close its page and then comeback to the experiment page. Once there, click on the next Java code example to review.

Lastly, after reviewing all code examples and answering their questionnaires, please answer our debriefing questionnaire.

TASK EXPECTATIONS

We expect the following throughout the task:

1. We expect you to review and understand a total of 3 Java code examples, in the order provided in this page.

2. We expect the whole task be done within 2 hrs; including the the time spent reviewing this page.

3. We expect the subjects to use only the resources available on the question-answer page to complete the task; i.e., no external resources like Wikipedia should be used.

LET’S GET STARTED

So without further ado, please click on the first link of Code Example 1 below to start the task:

1. Code Example 1 (src, q&a)
2. Code Example 2 (src, q&a)
3. Code Example 3 (src, q&a)

Stircup

2015-01-01T00:00:00+00:00

Stircup: Search application for (StackOverflow) Code examples

A brief introduction to get started with Stircup.

Its easy to get started with it. You use it by first entering a query, and then pressing the Search button below. Behind the scenes, the application fetches a few batches of code examples from StackOverflow ranked by significance to a query, ignores those ones that don’t match some user given criteria, and then displays the top k most relevant ones.

The criteria we are using include: code examples contain only Java code, at least 5 lines of code. The final result is a list of k most relevant (and or typical if the sorting by typicality checkbox is checked) code examples.

The first time you use it, the results get cached on your browser. After that, every time you search, this is done on your computer.

Code examples sorted by typicality

Thank you for trying this app.

Violette

2014-07-22T00:00:00+00:00

Code modification ideas in-place

Violette is Vesperin’s companion Chrome extension.

When installed, Violette detects Java code blocks on StackOverflow and then transforms them into a form that allows developers to actually edit and change the code in the web page of the Q&A system (in-place).

This in-place editing is done either manually via direct editing or semi-automatically via Vesperin’s built-in operations (available via the Kiwi API).

Out of the box, Violette supports the following operations.

Programming language detection using heuristics.
Code transformations and refactorings.
Content annotation via notes.
Lightweight drafts management.
Syntax correctness verification.
Code examples multi staging
Code examples summarization
Preview mode.

Violette’s latest version is available here.

Violette’s source code is available here.

Additional information

All the curated code snippets will be available, as tweets, on Twitter. If you want to see them, please follow @codedetour.

Kiwi

2014-07-22T00:00:00+00:00

Curation and Parsing services

The Kiwi API was implemented in Scala. It consists of two logical services: Curation and Parsing.

The Curation service manages Vesperin’s curation operations, such as basic code transformations and refactorings, code presentation, and publication (published on Twitter). The Parsing service manages the access to a Java parsing API, built on top of Eclipse JDT¹.

All requests are, by default, provided as JSON. There is no authentication required to make Curation and Parsing API calls.

Kiwi HTTP layer uses the Spray² framework to build a REST/HTTP-based integration layer. This layer is responsible for serving HTTP requests. It has the following HTTP endpoints as an API:

POST /kiwi/eval? {curation command}
PUT /kiwi/eval? {persist code command}
GET /kiwi/find?q={query text}
GET /kiwi/render?id={code id}

Here is an example of a curation command (e.g., Rename method):

{"rename": {
  "what": "method",
  "where": [1, 6],
  "source": {
    "name": "...",
    "description": "...",
    "content": "..."
  }
 }
}

Kiwi’s source code is still private. However, once it becomes public, it will be available here.

If you want to get access to this code, please send me an email.

http://www.eclipse.org/jdt/ ↩
http://spray.io/ ↩

Vesper

2014-06-22T00:00:00+00:00

Vesper: Code transformation utilities

Vesper is Vesperin’s curation library for parsing, manipulating, and reconstructing Java code. Vesper is built atop Eclipse’s JDT Java Parser and AST/DOM transformation utilities. This mechanism allows Vesperin to work with partial and non-compiling programs, not just complete ones.

The following code shows how one can start using the Vesper library.

import static edu.ucsc.refactor.ChangeRequest.forIssue;
import static edu.ucsc.refactor.ChangeRequest.reformatSource;

final Source  code  = new Source(...);
// using your own configuration:
// refactorer = Vesper.createRefactorer(new MyConfiguration());
// using default configuration
final Refactorer  refactorer = Vesper.createRefactorer();

final Map<String, Parameter>  userInput = ...;

// I. Ask the Refactorer to recommend changes for you

// print the reason for the change
final Introspector introspector = Vesper.createIntrospector();
final Set<Issue>  issues = introspector.detectIssues(code);
final List<Change>  changes = Recommender.recommendChanges(
	refactorer, 
	code, 
	issues
);

for(Change each : changes){
   System.out.println(each.getCause().getName());
}

// perform a single recommended change

final Commit applied = refactor.apply(changes.get(0));
System.out.println(applied.more());

// or handle All recommended changes

List<Change>  recommended = Recommender.recommendChanges(
	refactorer, 
	code, 
	issues
);

while(!recommended.isEmpty()){
    final Commit applied = refactor.apply(recommended.get(0));
    System.out.println(applied.more());
    final Introspector introspector = Vesper.createIntrospector();
    final Set  badStuff = introspector.detectIssues(code);
    recommended = Recommender.recommendChanges(
			refactorer, 
			code, 
			badStuff
		); // get an updated list of changes
}


// II. Dealing with random edits started by the user

// reformat Source's content

Change reformat = refactorer.createChange(reformatSource(code));
final Commit applied = refactor.apply(reformat);
System.out.println(applied.more());
  

Vesper’s code is available here.

Smalls chunks of text in small chunks of time

2013-10-08T00:00:00+00:00

I have a hard time sitting still and on track. I get easily derailed by many things, such as social media, news, a wandering mind, worries, etc.

Consequently, I’ve tailored my writing routine around this restlessness.

I live by my timers, writing in chunks until time is up. Then I take a few breaks. I used these time periods to quell any distractions.

I give myself 30 minutes to write.

I follow the chunking method. This method makes reading and understanding faster and easier.

I do my best to maintain this routine since it improves my focus and concentration.

It also helps me convey information more efficiently. I don’t get burned out. And there is no chance for social media to derail me.

I also apply this method when writing code; e.g. writing small chunks of functionality in small chunks of time.

Cyberbullying detection in social networks

2012-04-29T00:00:00+00:00

According to Google Analytics and to other sources, my data mining (ISM 245 class) project titled “Twitter Bullying Detection” is getting some attention. An earlier version of the project report was hosted on the co-author of the project’s website. This version is good in terms of communicating the project’s main idea and motivation. However, I think it is always important to have the latest version available. Therefore, I decided to post this latest version here.

Please feel free to contact me if you have any questions about the content of that report. Thanks.

Removing duplicate App icons in Launchpad (OS X 10.6+)

2012-04-20T00:00:00+00:00

If there are multiple icon instances of one application in your Mac’s Launchpad, then here is how to get rid of those duplicate instances:

Delete, or you simply drag and drop the application from Launchpad to the trash. Lion OS will automatically remove all the icons in Launchpad related to that App.
Then, simply drag and drop the App from the trash back to the applications folder. You will notice now there is only one icon for that App in your Launchpad.

I am hoping Apple will solve this problem soon.

Update

OS X 10.6 or later does not have this problem anymore.

Hello World!

2011-11-05T00:00:00+00:00

Here is the first post on huascarsanchez.com - my new website. My first blog post is the famous Hello World! message. Tell me, what’s better than the famous Hello World! message to introduce a new blog? ;-)

#!/usr/bin/ruby

# Hello world ruby program
puts "Hello World!";

Huascar Sanchez Posts

Publications

Curriculum Vitae

Huascar Sanchez

Read to Understand (RTU) Task

PREREQS

GUIDELINES FOR THE READ TO UNDERSTAND

GOAL

GUIDELINES

THE TASK

TASK EXPECTATIONS

LET’S GET STARTED

Multistaging to Understand (MTU) Task

PREREQS

GUIDELINES FOR THE MULTI STAGING TO UNDERSTAND

GOAL

GUIDELINES

THE TASK

TASK EXPECTATIONS

LET’S GET STARTED

Data

Data set

MTU/RTU answers

Source Code Curation Tooling for the Code Forager

Abstract

Codepacking

Codepacking

My book is out!

Software Patterns, Knowledge Maps, and Domain Analysis

Violette with Multi Staging and Code Summarization

PREREQS

THE TASK

TASK EXPECTATIONS

LET’S GET STARTED

Multistager

Code Multistager, Multistaging to Understand

Multi Staging to Understand Task

PREREQS

GUIDELINES FOR THE MULTI STAGING TO UNDERSTAND

GOAL

GUIDELINES FOR MULTI STAGING2

THE TASK

TASK EXPECTATIONS

LET’S GET STARTED

Read to Understand Task

PREREQS

GUIDELINES FOR THE READ TO UNDERSTAND

GOAL

GUIDELINES

THE TASK

TASK EXPECTATIONS

LET’S GET STARTED

Stircup

Stircup: Search application for (StackOverflow) Code examples

SEARCH

Code examples sorted by typicality

Violette

Code modification ideas in-place

Additional information

Kiwi

Curation and Parsing services

Vesper

Vesper: Code transformation utilities

Smalls chunks of text in small chunks of time

Cyberbullying detection in social networks

Removing duplicate App icons in Launchpad (OS X 10.6+)

Hello World!

GUIDELINES FOR MULTI STAGING²