Projects
Tech stack: LangGraph, Google Gemini, FastAPI, Python, TypeScript, Supabase, Silero VAD, Apify, Amazon SQS, AWS ECS, Amazon ECR, GitHub Actions, Vercel
- Built a multimodal recipe extraction pipeline that turns short-form cooking reels into structured ingredient lists and steps by fusing speech (including Burmese), on-screen captions, and visual frame understanding.
- Orchestrated ingestion and generation with LangGraph, Google Gemini, and Silero VAD, tagging each ingredient and step with its source (audio, caption, or visual) for transparent recipe cards.
- Shipped a full-stack product with a FastAPI backend, Supabase auth/storage, and a TypeScript frontend, deployed to Vercel.
- Designed an async jobs + worker system where the API enqueues job IDs on Amazon SQS and ECS workers claim, run, and complete ingest/recipe/audio/visual pipelines without blocking requests.
- Automated CI/CD with GitHub Actions: Docker images are built and pushed to Amazon ECR, then the API and worker services are deployed to Amazon ECS on successful main-branch builds.
HowToKeepPlantsAlive: AI Plant Recommendation System & Care Platform
Tech stack: LangGraph, PyTorch, Python, MongoDB, Gemini, Cohere, DVC, Prefect, MLFlow, Pinecone, Feast, TypeScript
- Engineered a two-tower recommendation system that recommends plants based on users’ environment and preferences, improving Recall@10 by 6.6× over a semantic similarity baseline.
- Designed a rule-based simulation framework to generate synthetic plant–user interaction data and benchmark the recommendation model’s ability to learn behavioral patterns, successfully capturing 8 of 10 predefined rules.
- Developed LangGraph AI agents capable of plant Q&A and comparisons, using a hybrid RAG pipeline of Pinecone vector search and BM25 retrieval over curated botany datasets.
Chico State University Chatbot: DxHub AI Summer Camp at Cal Poly San Luis Obispo
Tech stack: Node.js, AWS Bedrock, AWS Lambda, AWS API Gateway, AWS OpenSearch Serverless, AWS S3, AWS CloudFront
- Selected as 1 of 100 participants from 1,000+ CSU applicants for the DxHub AI Summer Camp and assigned to a team developing an AI chatbot for Chico State University stakeholders.
- Engineered a production Retrieval-Augmented Generation (RAG) system within the AWS ecosystem by building a knowledge base from 230+ university and local business webpages, enabling transparent responses with inline citations and source references.
- Developed a serverless backend using AWS Lambda and API Gateway to orchestrate user requests and perform LLM inference with Amazon Bedrock Claude Sonnet, supporting English/Spanish queries for 500+ daily users.
GaitorGate: Full-Stack Search Engine for AI Tools
Team Lead + Backend/Database Engineer
Tech stack: LAMP stack, Flask, AWS EC2, Google Gemini
- Led a team of 6 over a 4-month period to develop and deploy a production-grade search platform for 200+ AI tools, winning 1st place out of 12 teams in a software engineering competition.
- Architected a 14-table relational MySQL database and search pipeline supporting ~190 AI tools and 500+ keywords, implementing FULLTEXT and foreign-key indexes for efficient search and data retrieval.
- Developed Flask API endpoints for authentication, query execution, and user-uploaded tool ingestion, validating system reliability through 20+ student testers.
Board2Board: Chess Utility Tool for Over-the-Board (OTB) Game Recognition
Tech stack: Keras, Python, OpenCV, ResNet50, Scikit-learn, Matplotlib, SciPy, scikit-image, TensorFlow
- Designed a computer vision pipeline with OpenCV to detect, warp, and segment real-life chessboard images into 64 cropped squares for piece classification.
- Fine-tuned a ResNet50 model on a custom dataset of 2,000+ images, achieving 94% classification accuracy across all classes.
- Developed an adaptive thresholding system using linear regression to handle variable lighting and image conditions, improving successful board detection rate by 4× across 150 diverse real-world unseen board images.