At Coreworks AI, I owned part of the multi-agent harness and the correctness of its output for a product that turns business data into traceable reports, decks, and spreadsheet models. I built the spreadsheet agent, which creates models through tool calls and checks each formula and table before acceptance.
I also set up the Langfuse evaluation framework using test cases from real production failures. I added guardrails and monitored latency and misclassification rates.
At Aidora Inc (via Multiplier), I built a crawler and RAG pipeline from Scrapy crawling through Milvus vector search, along with its golden dataset and evaluation system. I also built a multi-agent system connecting LangGraph agents to internal backend APIs through MCP.
My work has included building a Neo4j knowledge graph of Indian law, training and deploying document models, and developing churn, customer lifetime value, and recommendation pipelines at Tata Digital (Tata CLiQ). I also built Shrota, a Hindi AI phone-survey platform, as a sole engineer.
