Member of Technical Staff - AI Memory Engineer

Member of Technical Staff - AI Memory Engineer

San Francisco Bay Area

·

Full-time

level dependent

Build how Nirva remembers a person over time, from everyday moments to lasting preferences and relationships. An AI engineering role for someone who makes personal memory accurate, retrievable, updatable, and firmly under the control of the person it belongs to.

← BACK TO CAREERS

Member of Technical Staff - AI Memory Engineer

San Francisco Bay Area

·

Full-time

·

180k-350k

·

level dependent

ABOUT NIRVA

Nirva is building the first AI fashion accessory for your mind — a necklace or bracelet that privately listens only to the wearer to enable proactive journaling, emotion tracking, and life coaching. Think Oura Ring for your Mind.

Backed by Lightspeed and South Park Commons, recognized at CES within 90 days of founding, and launching in the U.S. this fall. Co-founders bring backgrounds from Meta Reality Labs (Quest, Ray-Ban glasses) and fashion industry.

THE ROLE

We’re looking for an AI Memory Engineer to build how Nirva remembers a person over time: the memory system at the heart of our personal companion. Nirva is not a chatbot with a longer context window. It listens to real life, and it has to turn messy transcripts, events, and moments into memory that is accurate, updatable, retrievable at the right moment, and controllable by the user. You will design what gets remembered, how it is structured, summarized, consolidated, corrected, and forgotten, and how it is injected into chat, journaling, recaps, and proactive experiences so the product feels genuinely personal, not generic. You will work directly with the founding team to build this system from the ground up.

WHAT YOU'LL OWN

• Design and build Nirva’s long-term memory system for a personal companion: episodic memory (what happened), semantic memory (facts, people, preferences, relationships), and evolving user models over weeks and months

• Build extraction pipelines that turn transcripts, user context, events, and interactions into structured, high-quality memories, with source, confidence, and time attached to every memory

• Build consolidation, summarization, and hierarchy: how raw moments become daily summaries, life chapters, people profiles, and durable preferences without losing the specific detail that makes them feel true

• Own retrieval and context injection: how Nirva retrieves, ranks, summarizes, and injects the right memories into a prompt or agent workflow, within a context budget, so responses feel specific and grounded

• Handle updating, correction, conflict, and forgetting: new information replaces old information, contradictions are resolved, stale memories decay or are archived, and no hallucinated memory becomes “fact”

• Build user trust and control: users can see what Nirva remembers, edit it, delete it, and understand why it was used

• Build evals for memory: recall and precision of retrieval, stale and hallucinated memory rates, longitudinal consistency, and failure analysis on real, messy personal data

• Partner with product, design, audio, and recommendation teams so memory powers chat, journaling, daily recaps, emotional insights, and proactive experiences

WHAT WE'RE LOOKING FOR

Must-haves

• 2+ years of full-time engineering experience, or a PhD in computer science, machine learning, NLP, information retrieval, or a related field, plus 2 applied internships (industry or applied research) where you shipped or evaluated real systems

• Hands-on experience building a memory system, personal knowledge system, or long-term personalization system, not just a one-shot RAG demo: extraction, storage, retrieval, ranking, and updating over time

• Strong working knowledge of LLMs, embeddings, retrieval, structured outputs, summarization, classification, and context engineering, and how each fails in production

• Experience structuring personal data: knowledge graphs, entity and relationship models, timelines, or context stores

• Strong eval and debugging mindset for AI systems: you can define what a good memory is, build test cases from real failures, and improve extraction, retrieval, or prompts systematically

• Care for accuracy and trust in personal data: provenance, confidence, privacy, and user control are part of your design, not an afterthought

• Strong product engineering instincts and high ownership: you turn model capability into polished, user-facing experiences and drive work to shipped quality

• Comfortable with ambiguous, subjective, human-centered AI problems, and energized by 0-to-1 work, fast iteration, and early-stage startup intensity

• Based in San Francisco and able to work in person

Strong signals (nice-to-haves)

• Built memory for an AI companion, journaling, coaching, wellness, or consumer AI product

• Experience with long-term user modeling, personal knowledge graphs, or life-logging data

• Eval frameworks, data flywheels, fine-tuning, or human feedback loops for memory quality

• Voice / audio context: building memory from transcripts, diarization, or real-world events

• Experience spanning backend, AI infrastructure, and user-facing surfaces, including mobile

COMPENSATION

San Francisco Bay Area · Full-time · 180k-350k, level dependent

TO APPLY

Apply here