Engineer empathetic, secure LLM orchestration pipelines grounded in clinical psychology and zero-knowledge privacy.
Role Overview
ReAlign’s AI Companion, Atlas, is not a generic chatbot. It is a psychologically structured reflection system guided by CBT, Behavioral Activation, and strict safety guardrails. You will architect multi-provider AI gateway cascades, prompt injection defenses, and semantic vector retrieval over encrypted embeddings.
What You'll Do
Build and refine our multi-tier LLM gateway cascade across Google Gemini Flash, OpenRouter, and deterministic fallbacks.
Architect sub-5ms heuristic crisis intercepts that immediately bypass AI during acute distress to surface certified crisis resources.
Develop high-dimensional semantic search and memory graphs over encrypted pgvector stores.
Implement state-of-the-art defenses against prompt injection, canary token leakage, and markdown exfiltration.
What We Look For
3+ years building production AI systems, LLM evaluation harnesses, and vector search pipelines.
Strong fluency in TypeScript/Node.js, PostgreSQL/pgvector, and modern AI provider APIs.
Thorough understanding of safety alignment, latency optimization, and semantic caching.
Firm conviction in user privacy: data must never be shared, harvested, or leaked for training.
Nice to Have
Familiarity with RFC 5869 HKDF key derivation and envelope encryption patterns.
Background in affective computing, natural language sentiment analysis, or computational psychology.
Apply for this role
We value thoughtful writing over credentials. Send your CV, portfolio or code samples, and a brief note about why ReAlign resonates with you.