Head Of AI Engineering

The Role We’re looking for a Head of AI Engineering to own the technical direction of Thout.ai’s core AI systems end-to-end: the multi-pass LLM pipelines, agentic RAG architecture, structured extraction systems, and the model serving infrastructure underneath all of it. This is a hands-on leadership role, not a purely managerial one. At this stage, you are the most senior technical voice on AI: writing code, debugging production issues, and making architecture calls in the same week you’re setting technical roadmap and hiring the team to execute it. This role rewards a bias to action, since specs and priorities will shift weekly and the expectation is to build anyway. You’ll report directly to the CTO and work closely with the founding team on product direction, given the stage of the company. Within your first year, you’ll also be building and leading a small AI engineering team. Team today: You will be the founding AI engineering hire, working directly with the CTO and co-founders Team you’ll build: AI/ML engineers, applied research, and eventually MLOps, as the company scales What You’ll Own Technical strategy & architecture Set the technical direction for Thout.ai’s core AI pipeline: transcript processing, structured TLDR generation, action item extraction, PII detection/masking, and daily briefing generation Own architecture decisions for agentic, tool-using LLM systems (ReAct-style planning, MCP- based orchestration) with production-grade guardrails, evals, and observability Design and defend build-vs-buy and model-selection tradeoffs (open-source vs. proprietary, fine-tuned vs. base) against real cost, latency, and accuracy constraints Hands-on system building Design and debug multi-pass LLM pipelines, including subtle production issues like prompt- cache invalidation from schema injection, and architect around them (stable tool schemas, client-side structured-output validation, etc.) Build and maintain hybrid retrieval systems (BM25 + dense, re-ranking, query rewriting) tuned for meeting transcript dataOwn memory management and context layer architecture: how conversational history, entities, and prior meeting context persist and get surfaced across sessions, balancing context window limits against retrieval cost and latency Build long-context and cross-session memory systems that let the product connect ideas across meetings over time, not just within a single transcript Own model serving and inference optimization (vLLM/TGI-class stacks, ensemble base + fine-tuned model strategies) to keep the product fast and affordable to run at scale Lead fine-tuning efforts for domain-specific extraction tasks (e.g., PII/NER systems) as a named product vertical Team & process Hire and mentor a founding AI engineering team as the company scales Establish internal engineering standards: eval pipelines, prompt style guides, structured output conventions, and configuration practices that the team can build on consistently Balance startup speed with the rigor needed for a product handling sensitive client data (PII, confidential meeting content) Cross-functional partnership Work directly with the CTO and co-founders on product roadmap, translating business priorities (e.g., beauty industry client needs) into technical execution Represent the AI engineering function in strategic conversations, including due diligence, fundraising technical narratives, and client-facing technical credibility conversations as needed What We’re Looking For Master’s degree in Computer Science, Machine Learning, or a closely related field 8+ years of experience in AI/ML engineering, with demonstrated ownership of production LLM or NLP systems end-to-end, not just research or prototyping Deep, hands-on expertise in agentic RAG pipelines, tool-using LLM agents, and production- grade GenAI infrastructure Experience designing memory and context layer systems for LLM applications: long-context management, cross-session state, and context window/cost tradeoffs Strong systems fundamentals: model serving/inference optimization, fine-tuning (PEFT/LoRA- class techniques), and structured output enforcement at scale A track record of debugging non-obvious production issues (cache invalidation, schema drift, latency regressions) under real operating constraints Prior experience operating at a Lead or Principal level, ideally with some team-building or mentorship experience Comfort with ambiguity and speed inherent to an early-stage company: you’ll set your …