University of Central Florida — B.S. Computer Science
Expected graduation: December 2026 · GPA: 3.38
Milestones: Passed the Computer Science Foundation Exam
Funding: Bright Futures Academic Scholarship
Currently
Leading Knights Counsel through its pilot phase, with metadata schema implementation on the critical path. Running the Build-Recipe comparison design toward first results. Graduating December 2026 with plans on beginning Master's studies in the Spring. Looking for technical project lead and applied AI research roles.
Updated August 2026
Selected Work
All projectsTop 3 projects. Each page goes into detail about what I decided, what I rejected, and how I knew it worked.
Knights Counsel — permission-aware AI retrieval agent
An institutional-memory assistant for a university legal office, deployed inside the client's enterprise Microsoft 365 tenant. It answers from the office's own documents, cites its sources, and declines rather than speculating when retrieval does not supply enough grounding.
- Copilot Studio
- SharePoint Online
- Entra ID
- Power Automate
- React
- TypeScript
Build-Recipe Inference Agent — automated build inference
A containerized LLM agent that works out how to build software it has never seen: which build system, which dependencies, which commands in which order. Structured as a cross-tool comparison, because a reported success rate is only interpretable relative to the corpus it was measured on.
- Python
- Docker
- Anthropic API
- Linux
Harness in progress; comparison design complete, results pending.
MOREPocketProfessors — MERN collectible app
A full-stack app where users open packs and collect hand-drawn cards. I owned the pack-opening and collection views, the branding and card art, and the delivery plan.
- React
- TypeScript
- Node.js
- Express
- MongoDB
- JWT
Live at pocketprofessors.com
MOREWriting
All writingKnights Counsel: Technical White Paper, v1.0
Co-author · sole author, system design and behavior specification sections · August 2026
A full technical account of the system's architecture, behavior specification, evaluation methodology, and results, prepared for enterprise IT and the client. Available on request.
Knights Counsel: Final Design Document, v2
Contributing author · sole author, system design section · 2026
Requirements, architecture, and design rationale for the sponsored capstone, including the behavior test set and the architecture decision record covering the six-agent to two-agent reversal. Available on request.
Skills
- Languages
- Python, JavaScript, TypeScript, Java, PHP, C++, HTML, CSS, Bash, LaTeX
- Front-End & Web
- React, Node.js, Express, Fluent UI, SharePoint Framework (SPFx), semantic HTML, responsive CSS (Flexbox/Grid), WCAG accessibility patterns
- AI & Agent Engineering
- LLM agent design, retrieval-augmented generation, grounding and citation enforcement, guardrail and refusal design, tool-use and function-calling interfaces, structured JSON output contracts, Anthropic API, OpenAI API, Microsoft Copilot Studio, Microsoft Power Platform
- Evaluation & Research Methods
- Evaluation suite and rubric design, regression testing for non-deterministic systems, run-to-run variance analysis, pass@k, precision and recall, ground-truth construction, failure taxonomy, experimental design, literature synthesis, human-subjects study support
- Systems & Infrastructure
- Docker, Linux/WSL2, MongoDB, MySQL, Apache (LAMP), MERN, build systems (Autotools, Make, CMake, Maven, Gradle, Ant), SharePoint Online, Microsoft Entra ID, Power Automate
- Tools & Practice
- Git/GitHub, Jira, Confluence, Postman, VS Code, Overleaf, architecture decision records, technical documentation
Background
Undergraduate Research Assistant — Hardening Build Systems
– · University of Central Florida · Orlando, FL
- Contributed to empirical research on the security and reliability of open-source software build systems.
- Diagnosed build failures by hand across six build systems (Autotools, Make, CMake, Maven, Gradle, Ant), including undeclared JDK version requirements and compiler-flag incompatibilities that appear nowhere in the source tree.
- Built the tooling that converts raw experimental output into LaTeX-ready tables and pgfplots figures used in project reporting.
- Supported reproducible workflows by organizing datasets, validating results, and documenting analysis steps so runs could be repeated by others.
Research Assistant — Human Factors Group Study
– · University of Central Florida · Orlando, FL
- Ran experimental sessions for a study of how programmers debug code collaboratively in groups, contributing to data collection workflow and study setup.
- Instrumented participants with biometric equipment measuring heart rate, respiration, skin conductance, gaze, and prefrontal cortex blood oxygenation, improving measurement quality across sessions.
- Worked under human-subjects research protocol; completed the associated ethics training listed below.
Swim Instructor and Lifeguard — Rosen Aquatic and Fitness Center
– Present · Orlando, FL
- Teach small-group lessons across all ages and ability levels, from water survival through competitive swimming.
- Own class scheduling and documentation, large event setup, and pool surveillance.
- Communicate across age groups and language barriers in a safety-critical setting.
Awards & certifications
- Awards
- 3rd Place, UCF Horse Plinko Cyber Defense Competition (Oct 2024) · Bright Futures Academic Scholarship
- Research ethics
- Human Subjects Research: Social/Behavioral (Group 2) · Responsible Conduct of Research (Engineers)
- Technical
- Certified Internet Business Associate · MTA: Windows OS Fundamentals · MTA: Introduction to Programming Using Python
- Other
- American Red Cross Water Safety Instructor
About
I am a Computer Science senior at UCF who ended up specialising in the unglamorous half of AI work: deciding what a system must refuse to do, and building the evaluation that says whether it actually refused. That is the through-line from Knights Counsel, where the hard part was grounding and citation enforcement rather than the chat interface, to the Build-Recipe study, where the question is how much of any reported success rate is capability and how much is fit to the corpus it was measured on.
Before either, two years of research assistantships — hand-diagnosing build failures across six build systems, and instrumenting participants for a study of how programmers debug in groups. Outside of that I teach swimming, which is where most of what I know about explaining things to people who are not enjoying themselves comes from.
Email davidegusmao@outlook.com, or find me on LinkedIn and GitHub. There is a resume here too, and a PDF or CV if you would rather have the file.
