// Projects

Things I've built

End-to-end ML systems, MLOps pipelines, and domain-specific AI — all open-source on GitHub.

ChatArchive

Customer support chat analytics platform with an asynchronous ETL pipeline (FastAPI, Celery, Redis) that extracts raw conversation streams and queues them for processing without API lag, then loads spaCy-processed text and vector embeddings into PostgreSQL and Qdrant for semantic search.

PythonFastAPICeleryRedisPostgreSQLQdrantspaCyETL

GraphMind Agents

Multi-agent AI system using LangGraph and FastAPI that automates research and content generation workflows with human-in-the-loop approval, real-time WebSocket updates, and multi-LLM support (Groq, Claude, OpenAI, Ollama).

PythonLangGraphFastAPICeleryPostgreSQLRedisDockerWebSocketMulti-Agent

Pydantic-Recovery-Router

API validation gateway that intercepts unstructured AI and third-party outputs and validates them against strict Pydantic schemas before saving them into downstream databases, with a self-healing retry loop that reduced edge-case hallucination and serialization failures by 98%.

PythonPydanticData ValidationData QualityLLMStructured Output

Text2SQL MCP

Natural language to SQL agent built with FastAPI and MCP, enabling non-technical users to query a live sales database in plain English. Returns structured results with chart-ready outputs and enforces strict security safeguards.

PythonFastAPIMCPPostgreSQLNLPSecurityDocker

Vector Ingest Service

Asynchronous RAG pipeline using FastAPI, Celery, and ChromaDB for document ingestion, chunking, and semantic retrieval. Features background processing, persistent embedding storage, health monitoring, and structured validation.

PythonFastAPICeleryChromaDBRAGEmbeddingsDocker

VIGIL-FLOOD

Real-time flood monitoring platform built with Python and WebSockets, streaming live river water levels, rainfall, and sensor telemetry to public emergency dashboards with automated failover from damaged IoT sensors to satellite data feeds.

PythonWebSocketReal-Time StreamingIoTData IngestionFault Tolerance