Projects
Search or filter by type below.
A RAG-based job search platform: Next.js frontend, FastAPI backend, Qdrant + PostgreSQL data layer, and Groq (Llama 3.3-70B) for answer generation. Serves 100+ live postings with sub-1.5s response latency, and cut embedding memory footprint by 87% (400MB → 50MB) by switching to FastEmbed/ONNX to fit a 512MB free-tier host.
Real-time ASL recognition and translation web app. Rebuilt from a local desktop prototype into a production FastAPI + Next.js system with a WebSocket frame-streaming pipeline achieving under 100ms UI latency, plus an automated health-check and reconnection system for resilient uptime on free-tier hosting.
An AI productivity companion built during the Vibe2Ship Hackathon. React 19 + Vite frontend, with Groq's Llama 3.3-70B as the primary LLM and OpenRouter as an automatic fallback to stay resilient under rate limits.
A LangGraph agent (create_react_agent) with tool access via the Model Context Protocol, using Groq for inference, Tavily for web search, Qdrant for retrieval, and LangSmith for observability.
A parameterized data pipeline analyzing sentiment and engagement across AI writing tools (Jasper, Copy.ai, Writesonic, Rytr) using Reddit (PRAW), YouTube Data API v3, and Product Hunt's GraphQL API — scoped as a solo build.
No projects match that search.