AI Engineer

About Compute AI

Compute AI is building a personal AI agent for India and Bharat. Their first product, Jil, is a personal agent designed to manage administrative tasks like documents, payments, bookings, and follow-ups. Jil is built to run on edge devices owned by Indian users, prioritizing keeping personal data with the user and acting only with their consent. The underlying platform utilizes Ripple, their core algorithm. Compute AI is a small founding team with experience in building India's public digital infrastructure.

How we work

• Full-time, in-person role in Gurgaon.
• Emphasis on reliability, promptness, and ownership of mistakes with a focus on learning.
• Requires self-starters capable of identifying problems, proposing solutions, and driving them forward.

What you'll do

• Build the core platform for Compute and the Jil personal agent, which operates primarily on-device and uses larger models only when necessary.
• Select, fine-tune, distill, and quantize small language models (approximately 1–3B parameters) for reliable performance on entry-level Android hardware (around 4GB RAM).
• Implement and iterate on Ripple, the core algorithm, in collaboration with the chief scientist, transforming research ideas into measurable systems.
• Design the agent's planning, memory, tool usage, and permission-seeking mechanisms, with consent and reversibility as critical engineering requirements.
• Develop integrations with India's public digital infrastructure (identity, documents, payments, financial data) within consent and regulatory frameworks.
• Establish evaluation metrics for accuracy on real Indian tasks, hallucination rates, latency, battery, and memory consumption.
• Ensure personal data remains on-device and encrypted by default, with privacy as a core system constraint.
• Prototype rapidly, test with users, and be willing to discard unsuccessful approaches.

What you bring

• 0–5 years of experience building ML or AI systems, with an emphasis on shipped products.
• Hands-on experience in training or fine-tuning models, moving them from checkpoints to deployment.
• Deep knowledge of small language models.
• Thorough understanding of guardrails and their functionality.
• Strong fundamentals in algorithms and data structures.
• Experience with agent frameworks, tool use, function calling, retrieval, and memory, with insights into their limitations.
• Strong Python skills.
• Mobile (Android/Kotlin) or systems (Rust/C++) experience is a significant plus.
• Interest in following the AI conversation on platforms like X.

Bonus

• Experience with India Stack (UPI, Account Aggregator, DigiLocker, ABDM).
• Experience with multilingual or Indic language models.
• Experience with privacy-preserving ML.