Senior ML Engineer · Needl.ai · Mumbai
I'm a Senior ML Engineer at Needl.ai, where I build and own the agent platform and the LLM layer behind AskNeedl, Needl's answer engine for financial and enterprise data.
Over five years at Needl I've built every layer an answer passes through, from the connectors that bring documents in to the agents that now do the answering. These days I spend most of my time on agent harnesses, context engineering and evals, and on making an engineering team work well with AI coding agents.
Five years at Needl, one stack. I've built, in order, every layer a question passes through on its way to an answer. Pick a year, or a layer, to see what existed then and what I was building.
I joined Needl on the data side. The platform pulls in mail, drives, chat and the public web; I built the Microsoft Teams integration, a public-source crawler and the RSS pipeline, then hardened the Gmail, Google Drive, Outlook and Telegram connectors with resumable state, throttling and backoff.
Every answer starts here. If the document never arrives, nothing downstream matters.
I took over Needl's Elasticsearch search. First the search API moved from Node.js to Python; then proximity search, synonym and ticker expansion, spell correction, highlighting and multi-index querying over filings, earnings calls and regulator data, with a regression framework that checks results against golden responses.
The system learned to read. I added LLM-based query understanding (which companies, sources and dates a question is about), earnings-call summarization with sentence-level references mapped to PDF bounding boxes, the job service and table extraction behind it, and near-duplicate detection for news.
I joined AskNeedl, Needl's answer engine, and became its primary engineer: rebuilt the pipeline on Haystack, fine-tuned cross-encoder rerankers as the context filter, gated every model and prompt change on benchmarks, added web search and an answer-quality signal, and built its first multi-step agentic mode.
I turned the answer pipeline into Needl's agent platform: config-driven agents over a shared tool registry, skills that load context only when needed, tool-call budgets, per-fragment citations, an eval framework with LLM graders, and Langfuse tracing. Alongside it, the shared LLM layer over OpenAI, Azure OpenAI, Gemini and Claude, and text-to-SQL agents with input guards and read-only SQL safety.
One loop now runs sub-minute chat and twenty-minute due-diligence reports.
Before Needl: a B.Tech in Computer Science at IIT Dharwad (2017–2021) and Google Summer of Code 2019 with coala, which is where the writing below starts.
An eight-part log of my GSoC project with coala: reworking its action system so that analysis plugins (“bears”) can define their own fixes and suggest multiple alternative patches. Final report.