AI Reliability Engineer - Generative AI

About MantriAI: MantriAI is the search partner for this role. All applications come to us first. We screen and send only shortlisted profiles to the client. About the company: We are the GenAI video arm of a well-funded consumer entertainment group. We make video content using an Agentic AI pipeline - a story goes in, finished episodes come out. Over 25 shows are live, and we are scaling to thousands of hours of content. Total Experience: 2 - 6 years Why this role exists: A few months back, our pipeline ran on fixed workflows. We moved it to AI agents. Agents made us much faster, but they also do unexpected things - they take their own decisions, spawn sub-agents, and one small bad pattern can spoil many episodes. Today our product team debugs all this themselves. We want people whose full-time job is to make this system reliable, and then keep making it better. What you will do: - Read the traces of agent runs to find exactly which step, prompt or skill caused a bad output. - Study the good runs too. When an episode comes out unusually well, find what the agent did differently, and turn that into a permanent skill. - Rebuild fixes in our QA environment - change prompts, add skills, re-run - and prove with before/after numbers that your version is better. - Hand proven fixes to the dev team for production, then track them at scale. - Run your own experiments to make episodes better - stronger cliffhangers, consistent characters, better pacing. - Grow with the charter. We started with shows; promos and more are next. You should apply if: - You have built and shipped at least one AI project end to end, on your own, that solves a real problem. It is live or demo-able, and you can share the link. - When it broke, you found the exact cause and fixed it. You can walk us through that story. - You can explain why a change improved your output, not just that it did. - You build with AI tools daily - Claude Code, Cursor, Codex or similar. - You can work from our Bangalore office. Process: - 15-minute screening call with MantriAI - Take home assignment - 30-minute Live Assessment round with MantriAI - Interview rounds with the hiring company (preferably in-office) - Offer, usually within 2 days of the final round Posted by MantriAI, search partner on this mandate. Shortlisted candidates hear from us within 2 working days.