नमस्कार!I'm Sahil.about me ↓Full-stack software,AI where itcounts.
An engineer from Pune. By day I build backends and AI agents for founders and teams, and right now I'm building AgentDiff. By weekend I paint, cook, and explain how AI works on YouTube.
Available for new work · I reply within a day
Hi, I'm Sahil.
I'm an engineer from Pune. I did a five-year dual degree in Information Technology at IIIT Gwalior, and I've spent the years since building systems that have to hold: share allotments that can't be wrong, AI that enterprises have to trust.
These days I freelance for founders and teams, and I'm building AgentDiff, a tool that catches AI agents quietly getting worse.
I care about how things feel to use, not just whether they pass the tests. Away from the keyboard I paint, cook, and explain AI internals on YouTube.
Say hi: let's talk.
Currently
- Building AgentDiff
- Freelancing for First500days & JRat's Studio
- Writing The Latent Chronicle
IIIT Gwalior
Dual Degree (B.Tech + M.Tech), Information Technology
पुणे
Pune --:--:-- IST
Paint & cook
Built for production.
- १
Financial Engineering
A deterministic, scale-elastic processing platform for high-stakes IPO settlement and SEBI compliance. Engineered a vectorized 3-way reconciliation engine processing 1.6M+ records/sec under strict T+3 regulatory cycles.
Go · Python · PostgreSQL · AWS · Terraform · NextJS
- २
Intelligence Governance
The central governance and intelligence portal for the Omara ecosystem. Architected an enterprise-grade management hub that orchestrates multi-tenant workspace security and Ground Truth (GT) AI evaluation frameworks.
Python · Go · NextJS · Langchain · Langgraph · Neo4j
- ३
Developer Tools
Trajectory regression testing for AI agents. AgentDiff lines up a known-good agent run against the run from your pull request, and fails CI when the new path drifts, loops, costs more or recovers slower.
Python · GitHub Actions · LangGraph · CrewAI · OpenTelemetry · Langfuse
How I can help.
बुद्धी
AI engineering
Agents, GraphRAG pipelines and LLM evaluators, taken all the way to production.
- Agentic workflows (LangGraph)
- GraphRAG over your documents
- LLM evaluation & ground truth
- Production deployment
EngagementProject or contractStart a conversation
बुद्धी
AI engineering
Agents, GraphRAG pipelines and LLM evaluators, taken all the way to production.
- Agentic workflows (LangGraph)
- GraphRAG over your documents
- LLM evaluation & ground truth
- Production deployment
Where I've shipped.
Aug'26 - Present
Freelance
First500days
AI Engineer
Building the AI API layer for AYUVA, a UK healthcare navigation platform, covering symptom triage, medical report interpretation, care journey generation, and a guarded health assistant.
Building for AYUVA: AYUVA is a UK-based, AI-powered healthcare navigation and care coordination platform that helps patients organize records, understand their care journey, and prepare for appointments. Built NHS-ready with HIPAA/GDPR-grade security and patient-controlled consent.
Mar'26 - Present
Freelance
JRat’s Studio
Software Engineer
Developing secure, SEBI-compliant financial technology systems, optimizing high-performance data reconciliation pipelines, and architecting zero-trust AWS cloud infrastructure.
Building for Niche Technology Pvt Ltd: Niche Technologies Pvt. Ltd. is a SEBI-registered Category I Registrar to Issue and Share Transfer Agent (RTA) with over three decades of experience in India’s securities market.
Feb'26 - May'26
Freelance
Blimze
Design Consultant
Served as Design Consultant advising on visual strategy, crafting high-fidelity UI/UX mockups, and architecting interactive web application flows.
Oct'24 - Jan'26
Full-time
Omara Technologies
Founding Full-Stack AI Engineer
Founding Full-Stack AI Engineer architecting scale-elastic infrastructure, high-concurrency task distribution systems, and intelligent Neo4j/Gemini GraphRAG pipelines.
Tools of the trade.
बुद्धी
Gen AI & Machine Learning
भाषा
Languages
माहिती
Database
ढग
Cloud & DevOps
रूप
Frontend & UI/UX
7 of these shipped in the case studies. Hover one to see where.
First principles, on video.
Oct 2026
MHA vs GQA vs MQA: Attention Heads, Explained Simply
Your GPU has 80 GB of memory and the model only needs 14 GB. So why can it serve only about 30 people at once?
- Attention
- GPU Systems
- AgentDiff
- AI Agents
Next upFlashAttention · KV caching · Systems-level LLM optimisation
Have a hard problem? Let's talk.
Open to freelance & project-basis AI engineering, full-time & contract SDE roles, senior AI consulting and 1-on-1 mentorship. Remote, or onsite across India.