About Bash AI

Bash AI is building hiring intelligence: a platform that transforms the interview process between resume review and final offer. Given a role and a resume, it does the work a strong recruiting team would do with unlimited time: candidate and employer research, skill mapping, and structured question planning.

During interviews, Bash AI supports interviewers in real time with targeted questions, guidance on strong and weak answers, and focused nudges. After each conversation, it analyzes the transcript, validates claimed skills, and produces recommendations with evidence. Across rounds, it carries forward what it has learned, avoids repeat questions, and ranks finalists side by side.

We’re headquartered in the Bay Area, with our founder based there. The product is live with customers.

Our founding team has run hiring at scale. One founder built and exited an HR tech platform. Behind us sits a bench of advisors: serial founders and industry leaders who have built and sold companies in this market.

This isn’t a first attempt at a market we read about in a report.

The role

We’re ten people. Eight of them build the product. You’d be the one who owns the part that thinks.

Everything Bash AI does, from researching a candidate to nudging an interviewer mid-conversation to ranking finalists across four rounds, runs on agents. You’ll design them, build them, ship them, and fix them when they do something strange in production at 2 AM.

There’s no AI platform team above you and no internal framework you’re expected to follow.

If LangChain is the right tool, use it. If something newer is better, prove it and bring it in.

Most GenAI roles are prompt tuning on someone else’s architecture. This one is the architecture.

What you’ll actually do

Build agents that do real work. Multi-step reasoning, tool use, and memory that carries across interview rounds. Not demos. Systems that recruiters and hiring managers rely on every day.

Make them fast enough for a live conversation. An interviewer nudge that arrives after the candidate has finished answering is useless.



You’ll care about latency the way most people care about accuracy.

Make them right, and know when they’re not. Our recommendations come with evidence. That means you’ll build the evaluation that tells us whether an agent got better or just got different.

Work across the whole stack. With back-end engineers on how agents get data through our APIs and services. With front-end engineers on how agent output shows up in the product. Your work is only done when it reaches a user.

Keep an eye on what’s next. The agentic AI landscape changes monthly. You’ll evaluate new frameworks and models, and you’ll have the judgment to know which ones are worth switching for.

What we’re looking for

We’re an evidence company, so we care about what you’ve shipped more than where you’ve worked.

- Production Python. You ship code confidently, and you’ve maintained it after it went live.
- Real agentic experience. You’ve built with LangChain, LangGraph, or something similar, and at least one of your agents has been used by real users.
- You understand how LLMs actually work. Context windows, tokens, and the trade-offs between latency, cost, and quality. You know why a model fails, not just that it did.
- Retrieval experience. You’ve worked with vector databases or RAG systems and know where they break.
- You learn fast. You’ve picked up a recent framework under pressure and shipped with it.
- You work well across teams. You can explain an agent’s behavior to a front-end engineer and a founder in the same afternoon.
- Prompt engineering and agent evaluation experience is a real advantage, not a requirement.

Typically that adds up to 2-5 years of engineering experience, with at least 1 year on LLM systems.



If your experience is shorter but your shipped work says otherwise, apply anyway.

You should probably not apply if

- You want a clearly scoped ticket and a tech lead who reviews your architecture.
- Your agent experience is mostly tutorials, hackathons, or notebooks that never reached users.
- You’d rather research than ship.
- You’re uncomfortable with your work being measured.

What you get

- Ownership of the core. The agents are the product. You’ll own them.
- Direct founding team access. No layers. If you think the architecture should change, you’ll be talking to the people who can say yes.
- Real problems. Real-time reasoning, memory across rounds, evidence-backed recommendations, and evaluation of a task humans themselves disagree on.
- Early, not too early. The product is live and customers are using it. The architecture is still being shaped, and you’ll be shaping it.
- Equity in the company.

How we hire

We build interview intelligence. It would be strange to run a bad hiring process.

So: four conversations, all with people who make the decision. You’ll know where you stand after each one. We’ll tell you what we’re still unsure about instead of quietly disqualifying you for it. Nobody gets ghosted, including candidates we say no to.

1. Intro call (15 min). Your background, our story, and a quick look at the product.
2. The agent you shipped (30 min). Walk us through something you built and put in production. What it did, what broke, and what you’d do differently.
3. Live design session (30 min). We give you a real problem from our product. You design the agent with us. It’s not a memory test. We want to see how you think when the requirements change halfway through.
4. Final conversation (30 min). Ambition, expectations, comp, and everything you want to ask us.

To apply

Send us:

- Your résumé.
- A short note on an agent you shipped to production. What it did, what went wrong, and how you fixed it.
- One sentence on what you think LLMs still get wrong.

Bash AI is an equal opportunity employer. We welcome candidates from all backgrounds.

📌 GenAI Engineer (New Delhi)
🏢 Bash AI
📍 New Delhi