AI Engineering Intern — Voice Agents & LLM Evaluation (Python)

About the role

TheInterviews.ai runs real-time voice interviews: speech-to-text, an LLM interviewer that adapts follow-up questions, text-to-speech, and a scoring pipeline that has to be consistent and explainable. You will work on that brain, in Python, with the founder. Expect to touch prompts, evaluation harnesses, latency, and cost every week.

What you will do

• Improve the interviewer agent (Python 3.13, FastAPI, LiveKit Agents): question flow, follow-ups, turn-taking, and failure handling.
• Build and run evaluations: golden sets, score-stability checks, dimension drift guards, and regression gates that run in CI before a prompt or model change ships.
• Measure and reduce latency and cost across STT, LLM, and TTS; compare providers and models with data, not vibes.
• Read transcripts and scoring outputs critically, find failure modes, and turn them into tests.

Apply in 25 minutes

The fastest path is our own product: open the AI interview for this role, enter your email, follow the magic link, and take a 20-minute AI interview. That is round one. Or submit the form below with your resume and we will send you the same interview link.

Details

• Internship, approximately 12 weeks, starting October/November 2026. 20–40 hours per week, flexible, with a daily Pacific-time overlap window.
• Remote within the United States. Optional in-person days in San Diego.
• Students and recent graduates welcome, including F-1 CPT/OPT. Academic credit supported where your program allows.
• Meet us in person: ITServe Alliance job fair, San Diego, Tuesday 29 September 2026.

What we look for

• Strong Python, including async/await and writing testable modules.
• You have built something with an LLM API and can explain what went wrong and how you evaluated it.
• Comfort with basic statistics: you can reason about variance, sample size, and whether a change is real.
• Curiosity about speech and conversation systems; rigor about measurement.

Nice to have

• Experience with STT/TTS APIs, WebRTC or LiveKit, vector search, or fine-tuning.
• pytest, Docker, any cloud deployment.
• Coursework or projects in NLP, ML, or HCI.