InCommon
Website:
incommon.ai
Job details:
The question we're trying to answer
Since we started InCommon, one question has sat underneath everything: can hiring be deterministic? LLMs let us push the question to the extreme for the first time—make machines do everything they can do, layer human expert judgment on top, and see how far we get.
Hiring is one of the most important things any company does. It's where most CEOs spend their time, because the people you bring in determine whether you build something enduring or not.
This role is the engineering arm of that experiment—building the systems that let us test, at scale, how close to deterministic we can get without losing the judgment that makes a great hire actually great.
About us
Founded in 2023 and backed by Better Capital, InCommon helps global companies build exceptional teams in India. We obsess over transparency, authenticity, and delivering outstanding experiences—for both our clients and the talent we work with. We're driven by a mission to connect India's top talent with high-impact, rewarding careers.
You'd be our first engineer on the platform that makes all of this scale.
What you'll do
- Build the agent stack end-to-end. Sourcing, screening, interview copilots, fit evaluators. You own the harness—evals, rubrics, traces, regressions. Not just the prompts.
- Train and deploy small language models where latency, cost, or domain specificity actually matter. Know when a fine-tuned 3B beats a frontier call, and ship accordingly.
- Push the "what can machines do" frontier as far as it goes. Find the five steps begging to be automated and the one step where a human will always win. Ship the four.
- Sit with recruiters and clients. They're your first users, and the source of truth on what judgment actually looks like in this work.
- Help interview the next engineers. You'll be part of the bar.
- Move fast. Prototype Monday, in recruiters' hands by Thursday, iterate Friday.
Who you are
- 2–4 years of building. AI-native by instinct, not by adoption—you were using Claude Code and agent frameworks before your team asked you to.
- You've shipped at least one real agent system that went beyond a demo—tool use, memory, evals, the unglamorous plumbing.
- You've fine-tuned or trained SLMs and can talk credibly about tokenizers, LoRA, quantization, serving trade-offs. Not theoretically. You've done it.
- Solid software fundamentals. You care about latency budgets and failure modes.
- You can talk to non-engineers without translating. Recruiters will be your closest collaborators.
- The "can it be deterministic?" question genuinely interests you. You'll be living inside it.
Click on Apply to know more.