Professional networking has barely moved in twenty years. Voice infrastructure has moved even less — until now. Arrowhead is building autonomous voice AI agents that handle real-time calls at scale, across edtech, fintech, and healthcare. The belief behind the company is simple: solving for India's billion-person voice-first market solves for the world.

The clients are real and recognizable — PhysicsWallah, Vedantu, Aditya Birla, Paytm Insurance, Tata 1mg. The calls are live. The stakes on every interaction are high. Arrowhead is backed by leading investors at Seed stage, and the team is small, sharp, and moving fast.

This is a founding-team-adjacent seat. You'll work directly with the people who started this, which means the bar on rigor and autonomy is set at the top of the room from week one.

Built in Bengaluru. For a billion people. For the world.

This is Arrowhead's Prompt Engineer seat, though the work is closer to agent engineering than prompt writing. The company is already servicing a live book of customers across edtech, fintech, and healthcare — each with distinct call flows, edge cases, and outcome expectations. Every call can go sideways in a dozen different ways. Your job is to make sure it doesn't.

You'll sit at the intersection of what customers need and what the AI actually delivers on every call. That means deep customer conversations, prompt architecture, live call evals, and tight iteration cycles until the agent behaves exactly the way the business requires. The longer arc of this role is to build the rubric: a framework of use-case templates the entire team can pull from. You'll be working directly with the founding team, which means the bar on rigor, autonomy, and communication is founder-level from week one.

• The prompt architecture for live customer deployments. Every call flow across edtech, fintech, and healthcare runs on prompts you design, test, and own end to end.
• Customer discovery for each use case. You'll run the conversations that surface goals, edge cases, and outcome expectations — then translate them directly into agent behavior.
• Live call evals and iteration cycles. Run test calls with real users, identify where the agent breaks, and come back with specific semantic fixes — not general observations.
• The eval framework. Build the metrics and evaluation systems that tell the team, objectively, whether a call went the way it should.
• The use-case rubric. The longer arc of this role is a library of call-type templates the whole team can pull from — your most durable contribution to how Arrowhead scales.
• Prompt performance across model variants. Understand how different models affect outputs and own the decisions on which prompt structures hold across them.
• Founder-level reporting. You'll present findings, frameworks, and recommendations directly to the founding team — with the rigor and clarity that requires.

• You've done serious prompt or context engineering work in the last six to twelve months — at an AI company, on a product team shipping AI features, or as an AI PM who's lived at the customer-AI intersection. You know what prompt engineering actually means, and you've done it extensively.
• You think in frameworks and rubrics naturally. When you encounter a new class of problem, your instinct is to structure it, not just solve the instance in front of you.
• You've worked directly with customers to translate messy, real-world requirements into precise system behavior. The gap between what a customer says they want and what the AI needs to do is a gap you know how to close.
• You're a self-starter on two-way-door decisions. You run the experiment, check the data, and come back with a result — not a request for permission. You escalate one-way doors with a strong recommendation, not a question.
• You can point to the specific semantics in a prompt that explain why one call behaved differently from another. Not a general hypothesis — the exact line.
• You teach the people you work with. You bring your own research into the room, share what you're learning, and elevate the conversation rather than waiting for direction.
• You're genuinely obsessed with the problem. Building voice AI for a billion people in India is the kind of mission that either grabs you or it doesn't. If it does, that energy shows up in how you work.
• You understand how different models affect outputs and outcomes — not just in theory, but from having tested it with real prompts on real systems.

• Own the problem, not just the task. When you walk out of a room, you know exactly what you're going to do — because you asked the right questions before leaving.
• Think in frameworks. One-off solutions don't scale. Build the rubric, then apply it.
• Run the experiment. When data and customer feedback pull in different directions, test the hypothesis. Don't wait for consensus on a two-way door.
• Bring your research. Come to every conversation with something you've learned. Teach the people around you.
• Operate at founder-level rigor. You're working directly with the founding team. The bar on communication, precision, and follow-through is set at the top of the room.
• Zero fluff, high velocity. Sharp work delivered fast. If it can be tested in a day, it should be.
• Stay close to the call. The ground truth is in the audio. Spend time in live evals, not just in docs.

Seed-stage equity — 10 to 30 lakhs worth — at a company handling millions of live calls with recognizable clients. Cash compensation of 14 to 18 lakhs per year, with room to flex for the right person. A founding-team-adjacent seat, on-site in Bengaluru, building voice infrastructure for a billion people.