Vamsi K.

I build AI agents and data pipelines for teams whose data is complex, sensitive, and has to be right: financial models, filings, and research data that analysts currently dig through by hand. RECENT WORK I built a financial research platform for an emerging-markets advisory firm that was losing an estimated 40% of analyst time to manual Excel extraction. Analysts now ask questions in plain English through Claude and get answers in seconds, with a cell-level citation on every number. Delivered in 71 hours. Every value is validated byte-for-byte against the source workbooks, access is enforced per analyst at the data layer, and the firm has since onboarded external partners with scoped permissions. I later extended it with filings extraction and automated news monitoring with Claude-based significance scoring. WHAT I BUILD • MCP servers that connect Claude and other LLMs to your databases, spreadsheets, and internal tools, with row-level access control, OAuth 2.0/PKCE, and audit-friendly citations • Financial data pipelines: structure-aware Excel/PDF ingestion, validation suites, and normalized analytical stores (DuckDB, Parquet, Arrow, PostgreSQL) • AI agents and RAG systems for research, analysis, and reporting workflows, built to flag uncertainty and cite sources rather than guess • Self-hosted ETL and automation for teams that don't want to run Airflow HOW I WORK In finance, "probably right" isn't good enough. I treat correctness, traceability, and access control as part of the build, not extras after it. You get a working system, validation reports, and documentation your team can run without me. OPEN-SOURCE AND SIDE PROJECTS Dataflow (self-hosted ETL engine), db-mcp-suite (MCP servers for PostgreSQL, MySQL, and MSSQL with a semantic layer), and low-level systems work in Go and Java, including a custom Redis-style store and a UDP file-sync protocol. STACK Python, FastAPI, FastMCP, DuckDB, Apache Arrow, PostgreSQL, Claude and OpenAI APIs, LangChain, Docker, AWS. STARTING A PROJECT Send me a short description of your data and the questions you want answered. I'll reply with an approach and a rough timeline.