I built Ledger, a document RAG chatbot with hybrid BM25 and vector retrieval, RRF fusion, and cross-encoder reranking. I evaluated it with a RAGAS harness and added Redis caching and conversational query rewriting.
I generalized the chatbot into a multi-tenant MCP server, with per-client data isolation and ASGI-level authentication middleware. I also Dockerized the chatbot and server for deployment.
For ShopMCP, I architected an e-commerce MCP server exposing REST and MCP layers over Postgres and pgvector. I added a multi-round tool-calling gateway, Redis-backed cart with distributed locking, and middleware for authentication and rate limiting.
As an AI/ML Intern at Akiyam Solutions, I built flood-model orchestration, terrain preprocessing, and mesh-generation pipelines. I also contributed catastrophe-loss and climate-scenario modules, simulation validation, WebSocket streaming, and a pytest suite.
