Benjamin Barrera

Interactive AI Developer Intern · Tampa, FL

• Co-developed LangGraph recursive data-cleaning agent with tool backtracking and structured graph workflows with over 20 tools for complex multi-step dataset cleaning
• Researched and built a 23-function DataFrame evaluation framework to measure cleaning accuracy and detect uncleaned/failed rows and columns, enabling automated quality analysis
• Refined agent by improving prompts, adding uniform data sampling, and optimizing graph logic, achieving 52% token and 39% cost reduction (~379K → ~183K tokens) while maintaining cleaning accuracy

Software Engineer Intern · San Francisco, CA

• Architected AI agents in Next.js with MCP integrations to GitHub, MongoDB, and ClickHouse to query 100GB+ of company events data and generate charts/graphs via AI SDK
• Designed metadata-aware RAG pipeline using Chroma vector DB, reducing query context size by 9K tokens per call and lowering LLM compute costs

Software Engineer Intern · Philadelphia, PA

• Refactored legacy C# (.NET) RESTful API endpoints for wind turbine energy calculation and optimized response time by 9% via asynchronous calling, database indexing, and caching
• Engineered 45 unit test cases (xUnit), validating codebase changes and ensuring calculation reliability
• Configured build pipelines for containerized (Docker) .NET service with Azure DevOps CI/CD YAML and improved deployment time by 13% with concurrency jobs

Software Developer Intern · St. Petersburg, FL

• Built RAG app with React, Django (REST), and JWT auth that cut research paper surveying time by 17% using multi-shot GPT-4o with LangChain for semantic indexing and conversation memory
• Published paper on PyTorch multimodal models for verbal impairment detection using BERT, Wav2Vec, Openpose, and Transformer fusion architectures, achieving 72% accuracy

Undergraduate Research · University of South Florida

Semi-supervised graph-based approach to infer real-time air quality (PM2.5) in unmonitored regions using spatial data from OpenStreetMap and temporal data from PurpleAir sensors. Built a GRU + Graph Convolutional Network pipeline over a grid-based city representation of Dallas, TX, achieving RMSE 12.13 on held-out test data.

Emerging Researchers · NCUR Presentation · Board #86

Published Research · University of South Florida

Proposed a multilingual mild cognitive impairment (MCI) detection model using multimodal approaches analyzing acoustic and linguistic features. Outperforms existing ML models by identifying universal MCI indicators across languages, finding that speech duration and pauses are crucial markers in multilingual settings.

Google Scholar

Angular · Spring Boot · PostgreSQL · Supabase · Tailwind · Gemini API

• Constructed educational full-stack platform with Angular to teach planetology to K-12 with Gemini batch-generated learning and testing cards along with leaderboard display, winning 5th place
• Created Spring Boot backend with Supabase (PostgreSQL) and oversaw team of 2 to develop frontend

ReactJS · Express · Node.js · MySQL · Gemini API

• Spearheaded application using ReactJS, Express, and MySQL (raw queries) to automate medical coding translation for 100+ records of custom patient data using Gemini, achieving accuracy of 67%
• Collaborated with local clinic, improving medical code conversion speed by 84%

Technology Lead · Tampa, FL

• Coordinated tech curriculum and presented 4 workshops to over 150+ students teaching data structures & algorithms, backend (Django), frontend (React), and cloud deployment (AWS)

Bachelor of Science in Computer Science · GPA: 3.88
Coursework: Data Structures & Algorithms, Operating Systems, Software Engineering, Databases