AI Engineer

About
Compute AI is building a personal AI agent for India. And Bharat. Our first product, Jil, is a personal agent that works for the person who 'owns' it. It handles the small administrative weight of everyday life, like documents, payments, bookings and follow-ups, so people get time back for the parts of life that are not admin. It is built to run on the edge devices Indians actually own, to keep personal data with the person, and to act only with their consent. Underneath the platform is Ripple, our core algorithm.
We are a small founding team, working alongside people who helped build India's public digital infrastructure. We launch on Diwali 2026. The team will stay deliberately small, which means everyone who joins now shapes what this becomes.
This team has been in AI since 1994, built NowFloats (acquired by Reliance Jio), and has published 15 reports on 'Inside the Internet of India' via Digital Desh https://desh.digital/. Most of us (actually, the three of us) come from leading global organisations and universities and have a deep sense of purpose and agency.
How we work
Full-time, in person, Gurgaon, every day. This is a building-together role, not a remote one. No side-hustles.
Timings matter. Be on time and be prompt with replies, reviews and handovers. A small team runs on each person being reliable.
You will make mistakes. We expect them. What we care about is that you own them, say so early, and learn from them. We also expect you to learn from each other.
Self-starters only. Nobody will hand you a fully specified task list. You should be able to spot the problem, propose the fix and get it moving.
In return, you get a front-row seat (and ESOPs!) to how personal agents get built in India, at the moment the rules and rails are being set. You will have regular exposure to industry leaders, policymakers and builders across the AI and digital public infrastructure ecosystem.
What you'll do
Build the core of Compute (the platform) and Jil, a personal agent that runs primarily on-device and reaches out to larger models only when it must.
Select, fine-tune, distil and quantise small language models (roughly 1–3B parameters) so they perform reliably on entry-level Android hardware, around 4GB RAM.
Implement and iterate on Ripple, our core algorithm, with the chief scientist. You will turn research ideas into working, measurable systems.
Design how the agent plans, remembers, uses tools and asks for permission. Consent and reversibility are engineering requirements here, not UX afterthoughts.
Build integrations with India's public digital infrastructure (identity, documents, payments, financial data) within their consent and regulatory frameworks.
Set up evaluation: accuracy on real Indian tasks, hallucination on numbers and documents, latency, battery and memory cost. Treat a wrong EMI figure as a severe bug.
Keep personal data on-device and encrypted by default, and design every system with privacy as a hard constraint.
Prototype fast, test with real users, and throw things away without sentiment.
What you bring
0–5 years building ML or AI systems. We care more about what you've shipped than how long you've been doing it.
Hands-on experience training or fine-tuning models, not just calling APIs. You have taken a model from a checkpoint to something deployed.
Deep knowledge of small language models.
A thorough understanding of guardrails and how they work.
Strong fundamentals in algorithms and data structures. You can reason about why something works, not just that it does.
Experience with agent frameworks, tool use and function calling, retrieval, and memory. You also have opinions on where they break.
Strong Python. Mobile (Android/Kotlin) or systems (Rust/C++) experience is a significant plus.
You follow the AI conversation where it actually happens, on X and elsewhere.
Bonus: experience with India Stack (UPI, Account Aggregator, DigiLocker, ABDM); multilingual or Indic language models; privacy-preserving ML.
Talk to the CEO at jagriti@runcompute.ai
#howarewedoing

Work arrangement
No