Role OverviewWe are building an AI-powered personal-growth companion — a conversational agent that blends large language models with psychometric and personalization signals to have genuinely helpful, emotionally intelligent conversations. We are looking for an AI Engineer who is equal parts applied LLM specialist and pragmatic full-stack builder: someone who can design retrieval and memory systems, craft and evaluate prompts, and ship the production Node.js services that power them.You will own features end to end — from the model layer and data pipelines to the APIs the product runs on — and have a direct hand in how the assistant thinks, remembers, and responds.What You Will Do
- Design, build, and optimize LLM-powered features against a provider-agnostic client (OpenAI and Anthropic), including streaming responses, model routing, and cost/latency tuning.
- Build and improve our retrieval-augmented generation (RAG) and long-term memory systems — embeddings, vector search, semantic/episodic recall, intent classification, and context assembly. Own prompt engineering and evaluation: structure system prompts, design conversation modes, prevent regressions, and build lightweight evals to measure response quality and safety.
- Develop and maintain backend services and APIs in Node.js/Express, backed by PostgreSQL (with pgvector) and Redis.
- Build asynchronous pipelines (BullMQ workers) for summarization, memory extraction, embeddings generation, and scheduled jobs.
- Integrate real-time voice experiences (WebRTC + realtime speech APIs) and push notifications.
- Contribute to the React frontend where features span the full stack.
- Care about AI safety, privacy, and responsible design — especially in an emotionally sensitive domain — including crisis-handling logic and guardrails.
- Write clean, observable, production-grade code; participate in deploys (PM2/Nginx) and incident debugging.
Our Tech Stack
- AI / LLM: OpenAI (Responses, Embeddings, Realtime) and Anthropic (Messages) via a unified client; RAG with vector embeddings & cosine similarity; prompt engineering; semantic + episodic memory.
- Backend: Node.js, Express, PostgreSQL + pgvector, Redis, BullMQ, JWT/bcrypt auth, Helmet, rate limiting, node-cron, web-push.
- Frontend: React (CRA), React Router, Chart.js. Infra/Ops: PM2, Nginx, Linux, Git/GitHub.
- Real-time: WebRTC + realtime speech APIs.
Must Have Qualifications
- [3+] years building production software, with hands-on experience shipping LLM-powered features (designing the surrounding system, not just calling an API once).
- Strong Node.js / JavaScript and REST API design; solid SQL/PostgreSQL fundamentals.
- Practical experience with RAG, embeddings, and vector databases/search (pgvector, Pinecone, Weaviate, or similar).
- Demonstrated skill in prompt engineering and an instinct for evaluating and improving model output quality.
- Comfort with async/queue-based architectures (BullMQ, Redis, or equivalents) and caching. Sound judgment around AI safety, latency/cost trade-offs, and data privacy.
- Ability to work across the stack and own features independently.
Nice To Have
- Experience with multiple LLM providers and building provider-agnostic abstractions.
- Real-time / streaming experience (WebRTC, SSE, realtime voice).
- React frontend experience.
- Familiarity with evals / LLM observability tooling. Background or genuine interest in psychology, coaching, mental health, or behavior design. DevOps comfort (PM2, Nginx, Linux, CI/CD).
What Success Looks Like (First 3–6 Months)
- Ship measurable improvements to response quality, memory accuracy, or latency/cost.
- Strengthen our retrieval and evaluation pipelines so we can iterate on the model layer with confidence.
- Help raise the bar on safety and reliability in a sensitive, user-facing product.
Who You Are Curious, product-minded, and pragmatic. You can move fast without breaking trust, you sweat the details in user-facing AI behavior, and you care about building something that genuinely helps people.
Work Location: In person
