About

Position: AI / RAG Architect – Profile Intelligence & Trait Discovery Design and build the intelligence layer for the system that reads user profiles, retrieves contextually relevant information, and uses large-language-model reasoning to discover new traits, memories, and actionable intents in real time. Core Responsibilities 🔍 Retrieval & Understanding • Architect a Retrieval-Augmented Generation (RAG) pipeline that links structured and unstructured data inside user profiles, sub-profiles, proofs, and beacons. • Implement hybrid retrieval (BM25 + vector + reranker) tuned for user traits, history, and intent classification. • Design chunking, embeddings, metadata schema, and similarity metrics optimized for identity, lifestyle, and behavioral data. • Build pipelines to continuously update and re-index new user data, external events, and interactions. 🧠 Trait & Memory Intelligence • Use LLMs to extract traits, habits, intents, and actions from conversations and profile updates. • Develop unsupervised or semi-supervised models that detect new patterns or emerging interests (“latent trait discovery”). • Create a profile enrichment engine that merges traits across sub-profiles while preserving privacy and context isolation. • Implement confidence scoring, decay functions, and anomaly detection to keep traits current and verifiable. 📈 Analytics & Feedback • Instrument trace logging and telemetry to measure retrieval accuracy, trait-detection recall, and hallucination rate. • Build human-in-the-loop feedback dashboards so curators can confirm or reject new traits. • Continuously tune embedding models, rerankers, and LLM prompts based on behavioral analytics. Deep understanding of RAG architectures (vector + BM25 + cross-encoder rerankers). • Proven work with embeddings, semantic search, or LLM pipelines (LangChain, LlamaIndex, Haystack, custom). • Solid grounding in Python / Go / TypeScript, vector DBs (FAISS, Pinecone, Weaviate, pgvector), and Elastic/OpenSearch. • Familiarity with LLM API orchestration, prompt engineering, and context compression. • Experience turning LLM outputs into structured knowledge graphs or profiles. • Comfort working with PII, data-privacy, and policy-based redaction frameworks.