Ashtik Mahapatra

About

LLM/ML engineer who has shipped production systems in legal/financial-risk domains at Wolters Kluwer, including a hybrid OCR+deterministic+LLM extraction pipeline that structured UCC filings at massive scale and drove $6M+ in revenue. Also built LangGraph-based multi-agent “Deep Research” workflows with model routing, tool calls (MCP), persistence, and human-in-the-loop review, and partnered closely with policy writers to deliver LLM summarization that cut writing time by ~60%.

Experience

Education

Key Strengths

• Built and deployed production LLM+ML system extracting structured data from messy legal documents (scanned PDFs/free-form text)
• Designed cost/latency-aware hybrid pipeline (deterministic + OCR + LLM fallback with caching) to scale reliably
• Scaled entity resolution across millions of filings using blocking and fuzzy matching to balance recall vs false positives
• Delivered measurable business impact ($6M+ revenue attributed to production system)
• Built deterministic multi-agent workflows in LangGraph with bounded loops, state passing, failure handling, and human-in-the-loop review
• Strong evaluation discipline for agents (schema validity, tool-call success, hallucination rate, latency; golden sets + LLM judge + periodic human validation)
• Effective cross-functional collaboration with non-technical stakeholders to define quality criteria and create feedback loops for fine-tuning data
• Improved policy-writer productivity via LLM summarization (20–30% publishable; 50% with light edits; ~60% time reduction)