Things I've built and shipped, with source and demos.
CLI that parses a legacy SOAP/WSDL service once and generates a standalone gateway exposing every operation as REST, an MCP tool server, and an OpenAPI 3.0 spec, all from a single shared model so the three surfaces can't drift apart. Includes a WSDL-driven mock SOAP backend for zero-dependency testing and a GoReleaser + Homebrew release pipeline.
Go, SOAP, WSDL, MCP, OpenAPI, Cobra, GoReleaser
Telegram-bot journal where git itself is the database: CrewAI agents parse free-text notes into dated Markdown with YAML front-matter, generate daily/weekly/monthly summaries, answer natural-language questions via a hybrid deterministic + embedding-based RAG search, and self-heal missing or malformed entries, committing every change to git as the source of truth.
Python, CrewAI, Groq, Telegram Bot, GitPython, FAISS, RAG
Healthcare platform (550+ users) with multi-model cascade (Llama-4/3.3/3.1), analyzing 750+ reports. Retrieval-augmented (RAG) conversational AI using a FAISS vector index and Hugging Face (HF) embeddings; secure auth, session management, rate limiting (15/day), PDF parsing (up to 20MB) with Supabase. Won global Streamlit Connect Challenge; 100+ stars, 50+ forks on GitHub.
Python, Streamlit, Groq, Supabase, LangChain, FAISS, Llama
AI-powered health agent to analyze reports and deliver personalized diagnostics using Groq AI. Contextual embeddings via pgvector (vector search built into Postgres); secure JWT-based sessions with OAuth sign-in for privacy and usability. PDF extraction and validation pipeline handling up to 10MB reports with offline analysis fallback.
Next.js, Node.js, PostgreSQL, Docker, Hugging Face, Groq, NextAuth.js, pgvector
Atomic Design Pattern and MVC architecture. Real-time data and authentication with Firebase; Gemini API for one-tap message enhancement and queries. Integrated Cloudinary for media.
React Native, Expo, Firebase, Cloudinary, Babel, Gradle, Jest, Gemini API