I am a technical lead based in Pune, India, and my background is deliberately unfashionable for someone working in AI: distributed systems, data pipelines, and production support. I learned what reliability means by being the person who got paged.
The largest system I have architected processes 520 million telecom parameters every fifteen minutes on Golang, Kafka, Kubernetes and PostgreSQL. Not because throughput is impressive on its own, but because a fixed cycle time teaches you something a demo never will: a system that accepts more work than it can finish does not slow down, it collapses. That lesson turns out to be the single most useful thing I know about AI agents.
Since 2022 I have been building AI products — conversational analytics over sensitive multi-tenant data, multi-agent systems on CrewAI and Model Context Protocol, text-to-SQL engines, retrieval pipelines, an agentic voice assistant, autonomous lead-generation agents. Some of it worked immediately. More of it worked after I stopped trusting the model with things a model should not be trusted with.
What I noticed across all of it: the teams struggling with agents were rarely struggling with AI. They were missing validation at boundaries, bounded resources, traceable failures, and any way to tell whether yesterday's change made things better. That is ordinary engineering discipline, and it iswhat I now do for a living — infour domains I have actually shipped in.
On working independently: this is a deliberate choice, not a gap between jobs. Reliability work is the part of AI engineering that consistently gets deprioritised inside a company, because features have an owner arguing for them and reliability does not. Working from outside means I get handed the problem directly, with a scope and a deadline, and I am not competing with a roadmap for attention. I take a small number of engagements at a time for the same reason.
I also work at the business end of things, which is unusual for an engineer and occasionally useful: I have led an initiative analysing customer data to find expansion opportunities, built AI systems aimed directly at pipeline and retention, and presented enterprise positioning strategy to a CEO. I mention it because it means I will ask what a reliability problem costs you, not just how to fix it.
Based in Pune, India (+05:30). Local time —.
I hold afternoons and evenings IST open for calls, which covers European mornings and US mornings on the earlier side. Async-first either way.
pip install fast-intent-classifiernpm i ai-test-gen-angular200M configuration + 320M performance parameters per 15-minute ingestion cycle, on Golang, Kafka, Kubernetes and PostgreSQL.
A text-to-SQL agent serving 2,000 concurrent users with strict per-tenant isolation on HR data.
Conversational analytics replaced a manual reporting cycle that previously took two to three days.
Nine years across telecom, HR, recruitment and workflow automation — architecture through production support.
Updated
Worth reading before you hire anyone remote. These are commitments, not preferences.
You get a written findings document and a recorded walkthrough rather than a status meeting. Given the timezone gap that is not a compromise — it produces a better artefact, because writing forces the thinking to be finished.
If an estimate is slipping or an approach is not working, you hear it from me while there is still time to change course. I would rather deliver an uncomfortable update in week one than a surprise in week three.
If your gaps are ones your team can close, or the problem is not the one you think it is, I will say so. That has cost me engagements and earned me referrals, and I am not planning to stop.
Eval suites, traces, dashboards, documents. No dependency on me continuing, and nothing built in a way only I can maintain.
Nine years of history, recommendations and endorsements are on LinkedIn, and the code is on GitHub. For client references, ask me directly — I will put you in touch with people I have delivered for rather than sending you a curated quote.
Mar 2022 – present
Technical Lead
SourceFuse Technologies
Architecture and delivery across AI products and large-scale data platforms. Built AI applications on OpenAI, Claude, AWS Bedrock, LangChain, CrewAI and Model Context Protocol, and architected a configuration-driven telecom platform processing 520M parameters per 15-minute cycle. Led design reviews, mentoring, release management and performance work across several teams.
Feb 2021 – Feb 2022
Senior Software Engineer
SourceFuse Technologies
Cloud-native backend services on AWS — microservices in Node.js and Python on ECS Fargate, CI/CD with CodePipeline, technical designs and mentoring.
Dec 2018 – Jan 2021
Software Engineer
Flairlabs
Backend services and enterprise applications on Node.js, AWS and SQL Server. REST APIs, database design, and production support across the full lifecycle.
Jun 2017 – Dec 2018
Application Engineer
Merce Technologies
Backend features and microservices, heavy SQL work, and first exposure to running things in production — which is where most of what I know actually came from.
Not written up as case studies, but real and shipped. Happy to talk about any of them.
Candidate and job matching with NLP and vector embeddings, Elasticsearch search, resume parsing and ATS integrations, across employer and candidate portals.
PythonNode.jsElasticsearchBedrock
What makes this domain hardEnterprise CLM covering authoring, review, approvals, amendments and clause libraries.
AngularNode.jsPostgreSQLCKEditor
What makes this domain hardReal-time Zoom audio transcription with Whisper, then summarisation and action-item extraction.
WhisperBedrockFastAPIDocker
Specialised agents collaborating on story planning, character consistency, drafting and editorial review across a book-length manuscript. A serious exercise in long-horizon context management.
CrewAIMCPPython
Facial recognition over CCTV footage, using augmentation to improve recognition on poor-quality frames.
PyTorchDlibPython
Autonomous lead generation, WhatsApp retention workflows, and a HubSpot chatbot for qualification — AI applied to revenue rather than product features.
CrewAIHubSpotWhatsApp API
About an agent that is misbehaving, a role you think fits, or a system you would like a second opinion on. I answer specific questions about specific systems for free.
Hiring rather than contracting?
I will listen to the right role. Send the company, the role and the comp range and I will be straight with you either way.
Download my CVNothing matched. Try “cost”, “loop”, “evals”, or “rates”.