Skip to content
Search jobs inside your AI agent. Claude, ChatGPT, Claude Code, Codex, Gemini, Cursor and Grok
Connect

Recro

NewBe an early applicant

AI/ ML Engineer

Posted Updated 40 views

Description

This is a senior individual contributor role. You'll work as a peer to the other AI engineers on the team and own your problems completely: you choose the approach, build it, measure it, ship it and keep it working. If you've led a project or a small team before, that experience will help, because your work will often pull in backend, app and hardware engineers.

We're hiring from a core ML background on purpose. Much of the work ahead is model work: training and fine-tuning on data we're allowed to use, building evaluation sets, and deciding when our own model should replace a paid API. It is hands-on work, and you'll write and ship the code yourself.

What you'll own

  • Speech and speaker identity. Improve speech-to-text on real-world audio, separate speakers reliably, and recognize who is speaking across conversations, including when several known people are in the same conversation.
  • Memory and retrieval. Turn transcripts into memories that are accurate and grounded in what was said, with no invented people or facts. A line from the TV or a stranger at the next table should not become a fact about the user. When the user asks weeks later, the right memory should come back.
  • Agents and actions. Build tool-using agents on top of memory, for example reminders and follow-ups that act on what a user committed to in a conversation.
  • Our own models for cost and latency. Fine-tune or distill smaller models that match large API models on our tasks, and prove it on our evaluations before they ship.

Across all of these, you'll build the evaluation sets that tell us whether a change helped, and you'll keep tracking quality, latency and cost after launch. You'll start with one or two of these areas, chosen around your strengths and the product's needs, and take on more through the first year.

What end to end means here

  1. Define the problem with a metric and a baseline.
  2. Build or collect the data, within our privacy rules on what data a model may use.
  3. Train, fine-tune or choose the model, and run the experiments.
  4. Prove the change on an evaluation set we trust, including its cost and latency.
  5. Ship it into the product with the backend and app engineers.
  6. Watch it in production and fix what real users find.

Your first 90 days

  • Own one live problem, such as speaker identity or memory accuracy, with a written baseline and an evaluation set the team trusts.
  • Ship at least one measured improvement to production, with before and after numbers for quality, latency and cost.
  • Write a plan, with its data needs, for one place where our own model should replace an API call.

What we're looking for

  • Experience. 4 to 6 years of industry experience building ML systems, with a core ML foundation. You have trained, fine-tuned and evaluated models yourself, and you can explain why a model fails.
  • GenAI depth. About 2 years of hands-on work with transformers and generative AI. That could be fine-tuning language or speech models, building agents, voice AI, retrieval systems or similar work.
  • Shipped work. At least one model you took into a live product and owned after launch: the data, training, evaluation, serving and the fixes once users found the edge cases.
  • Ownership. You have taken a loosely defined problem, broken it down yourself and carried it into production.
  • Fundamentals. Solid probability, linear algebra and optimization. You can read a paper, implement it and tell us whether it helped on our data.
  • Engineering. Strong Python and PyTorch, and enough backend skill to put a model behind a service and debug it in production.
  • Education. A B.Tech, M.Tech, MS or dual degree from an IIT, NIT, BITS or IIIT Hyderabad. If you studied elsewhere and have exceptional public work, such as widely used open-source projects, apply and show us.
  • Clear writing. You can explain a trade-off to an engineer, a product manager and a founder.

Nice to have

  • Speech work: speech recognition, diarization, speaker verification or voice activity detection.
  • Experience leading a project or a small team.
  • Distillation, quantization, or serving models under tight latency and cost budgets.
  • Work with multilingual or code-mixed language data.
  • Papers or open-source work in speech, retrieval, long-term memory or agents.

How we work

  • On-site in Bengaluru. Device, app and AI engineers sit together, and many bugs need a real Neo and a real phone to reproduce.
  • We decide with evaluations and numbers. A change ships when it beats the baseline on quality without breaking our cost and latency budgets.
  • We use AI coding tools every day, and we expect you to as well.

What we offer

  • Competitive salary and ESOPs.
  • Ownership of problems that decide whether the product works for its users.
  • Unsolved problems in ambient speech and long-term memory, with room to train our own models.
  • Direct work with the CTO.



Requirements

Must haves:

- 4 to 6 years of experience with a core ML background. They should have trained, fine-tuned and evaluated models themselves

- About 1-2 years of hands-on GenAI work: fine-tuning LLMs or speech models, agents, voice AI or retrieval

- At least one model they took to production and owned after launch

- B.Tech, M.Tech or MS from an IIT, NIT, BITS or IIIT Hyderabad. Share someone from elsewhere only if their open-source work is exceptional and tell us why

- Bengaluru, on-site


It's a hands on IC role. Past lead experience is a plus.


Please avoid:

- PhDs, they'll be too senior for this role

- Backend engineers who have only called LLM APIs or built prompt wrappers

- Research profiles with nothing in production


Skills:- Training and Development, vLLM, Fine-tuning LLMs, Speech-to-Text (STT) and Inference Optimization

Skills

Apply