(3)About
I build AI systems that hold up in production: agents, assistants, retrieval and the ML behind them.
The longer version. I studied mathematics in Pune, then Data Science and AI at the University of Liverpool, where I still work as a machine learning engineer: a model that finds the direction of a radiation source from a single sensor, built with nuclear physicists. Somewhere in there the interesting problems stopped being about the model and started being about the product around it.
Today I build AI at Script Lanes, a medical e-learning platform, and at Nextoar, an industrial AI startup, and freelance alongside, on AI systems and on websites. Agents, retrieval, computer vision, the evals and tracing that keep them honest. A model is a component. The product is the thing. I'd rather ship something small that holds up than something impressive that doesn't.
Outside of work I'm learning German, reading Nietzsche's Thus Spoke Zarathustra and learning 3D printing, and I care about type nearly as much as tokens. If any of this sounds familiar, I'd like to hear from you.
Three rules I keep.
Design the seams
AI products fail where a confident system meets a human. I design those handoffs deliberately: confidence shown, sources cited, escape hatches everywhere.
Measure, don't vibe
Every project gets a small, honest eval set from real usage before we tune anything. Prompts change; the test set is the contract.
Write it down
Decisions, tradeoffs and the things we chose not to do live in the repo, not in someone's head. Future you is a stakeholder.
In numbers.
Year by year.
- Jun – Now · Role
AI Engineer Intern, Nextoar, Bangalore
Early-stage industrial AI startup, working directly with the founder and CTO. Agentic outreach with custom tool calling and MCP servers, computer vision fine-tuning for an industrial floor assistant, hybrid retrieval so troubleshooting answers surface exact part numbers, Langfuse tracing and model routing across providers.
- May – Now · Role
Software Engineer Intern, Script Lanes, Pune
Shipped a medical e-learning platform: a vision-model service that transcribes and classifies handwritten notes behind schema guardrails, a syllabus chatbot with classifier-routed retrieval over human-authored course material, cost-aware model routing, LLM-as-a-judge evals with medical student testers, and an AWS deployment with per-user private storage.
- Mar – Jun · Role
Apprenticeship, Constellation Marine Services, Dubai
Built and shipped a PERN-stack contact management system in under two weeks, replacing scattered Excel sheets holding 2,000+ client records: relational schema and migration pipeline, multi-field search, deduplication and validation, role-based access, CSV import and export. Lookups went from 15 minutes of spreadsheet searching to sub-second queries. Also wrote implementation-ready requirements for three internal workflow automations by shadowing non-technical stakeholders, and designed and built the Fujairah branch's public website.
Read the website case study - Jan – Now · Role
Freelancer, Pune
Taking on work directly with founders and small businesses, alongside the roles above: AI agents, retrieval systems and automations, and websites I design and build end to end.
- Jun – Now · Role
Machine Learning Engineer, University of Liverpool
Own the end-to-end pipeline for a model that determines radiation direction from a single sensor, built with nuclear physicists on leakage-free datasets. Git-versioned evaluation harness on a compute cluster; after Bayesian tuning of TensorFlow and PyTorch networks, shipped a simpler classical baseline that outperformed them.
- Education
M.Sc. Data Science & AI, University of Liverpool
Dissertation on classifying 3,207 breast tumours from bulk RNA-seq and finding the genes that drive the prediction. Graded A+.
Read the case study - Mar · Milestone
Persuasion Award, Data Quest 2024, Village Software
One-day data competition: six student teams mining a virtual island's resource grid to claim the three most valuable sites on a simulated market. With Mateus Padilha Luz as Byte Busters, built the models in Python and presented the result as revenue rather than charts. The judges gave us the Persuasion Award for it.
Village Software's write-up
- Education
B.Sc. Mathematics, Pune University
- Sep · Milestone
Runner-up, Avishkar Research Competition, Maharashtra
Undergraduate study of whether academic stream, sport and other extracurriculars predict how quickly someone solves the Tower of Hanoi. Surveyed 100 students, timed each on the puzzle and found no single factor was a strong predictor on its own. Second place statewide. Written up with Aditi Phadke and Rohan Uttekar and published in IJRAR, April 2026.
Read the paper
Want to talk? Say hello.
A project, a question, or just to compare notes. I read everything and write back within a couple of days.