AI / ML Engineer (AI Voice & Social Product) – w/ Equity
Bounty Amount: $20,000-22000
Company Name: Known
Role Type: Full-Time
Location: San Francisco, CA (onsite 5 days a week)
Salary / Hourly Rate: $200,000 - $220,000 per year
Benefits: US Citizens Only
Role Information
Role Overview: N/A
Responsibilities: Design and ship multi-stage retrieval + re-ranking for compatibility scoring, search, and personalization., Build and maintain data/feature pipelines for training, evaluation, and reporting; ensure reproducibility and data quality., Train, fine-tune, or prompt LLM/encoder models; manage model versioning, rollout, and rollback., Run offline evaluation (e.g., AUC, NDCG, MAP) and online experiments to measure real user impact., Stand up inference services with tight p95 latency and cost targets; add caching, batching, and fallback strategies., Implement safety/guardrails and monitoring for drift, bias, and failure modes; define model SLOs and alerts., Collaborate with infra/platform to productionize models and with product/design to turn signals from voice/text into better matches., Document decisions, write lightweight runbooks, and share dashboards that track match quality and model health.
Qualifications: 4 to 10+ years in infrastructure, platform, or data engineering with real ownership of uptime, performance, and security., Expert with AWS and Infrastructure-as-Code (Terraform, Pulumi, or CloudFormation)., Strong proficiency in Python or TypeScript, plus tooling/scripting (Bash/YAML)., Containers and orchestration experience (Docker, Kubernetes or ECS) and CI/CD pipelines you designed and ran., Proven ability to design and operate data pipelines and distributed systems for both batch and low-latency use cases., PostgreSQL at scale, ideally with pgvector/embeddings exposure for ML-adjacent workloads., Strong observability practices: metrics, tracing, alerting, incident management, and SLOs., Excellent collaboration with AI/ML and product teams; clear communication of tradeoffs and risk., Work authorization in the U.S. and willingness to be on-site five days a week in San Francisco., Experience supporting model training and inference pipelines, feature stores, or evaluation loops., Prior work with streaming voice, low-latency systems, or recommendation/retrieval stacks., Early infra/platform owner at a seed–Series B startup, scaling AWS with Terraform and CI/CD, Built real-time and batch data pipelines that powered matching, voice, or recommendations, Ran Postgres at scale (schema design, indexing, pooling), with pgvector or embeddings in prod, Set up observability and on-call (metrics, tracing, alerting) that improved SLOs, Partnered with ML to deploy and monitor model inference with clear latency and cost targets
Minimum Requirements: Must be authorized to work in the U.S. without future visa sponsorship.,Able to work onsite in San Francisco, CA five days per week.,3+ years in applied ML focused on ranking, recommendations, or search in production.,Strong Python; experience with PyTorch or TensorFlow (Hugging Face a plus).,Hands-on with embeddings and vector search (pgvector, FAISS, Pinecone, or Weaviate).,Proven experience taking models from notebook to production: packaging, APIs, CI/CD, canary/rollback, monitoring.,Data pipelines for training and evaluation (e.g., Airflow, Prefect, Dagster, or dbt) and sound data-quality checks.
Screening Questions: (Optional Video). This step is completely optional. If you’d like, record a short 2–3 minute video introducing yourself and your experience — or share a recording of your interview with the recruiter if that’s easier. You can upload the link via Loom or Google Drive. This just helps us get to know you better, but there’s no pressure if you’d prefer to skip it.,(Optional Portfolio / GitHub) If available, please share a link to your GitHub, portfolio, or any recent projects you’ve worked on. This is entirely optional but helps provide more context about your work.,What excites you about building matching systems at Known, and why are you considering a move now?,Matching/Retrieval Describe a ranking system you built. Include retrieval method, re-ranking approach (e.g., LLM or learning-to-rank), features used, offline metrics you tracked, and one online metric you moved.,Production & Reliability Tell us about one model you owned in production. Include throughput and latency targets, how you monitored it (drift, bias, alerting), and one incident you diagnosed and resolved with the before/after impact.
Company Information
About Company: N/A
Culture: N/A
Additional Information
Interview Process: Make sure the candidate has read the AI/ML one-pager: https://docs.google.com/document/d/1-NnpNV-q8FzYZSBX1rZzjM5ReCUAXgYf1NshB2K53Mo/edit?usp=sharing, Walk them through the product tech overview deck: https://www.figma.com/design/fMbQlGiJ0vPHoELqW1MzJA/Known-VC-Decks?node-id=292-5671, Confirm full working rights in the U.S., Confirm willingness to work 5 days per week on-site in the Marina office., Stage 1: Intro call — 30 minutes, Stage 2: Technical Screen I — 60 minutes15 min light system-design/ML discussion (matching, evaluation, tradeoffs)45 min Python coding exercise (problem-solving and clarity), 15 min light system-design/ML discussion (matching, evaluation, tradeoffs), 45 min Python coding exercise (problem-solving and clarity), Stage 3: On-site Technical II — ~3 to 4 hours~15 min briefing on the prompt~2.5 hours independent work building part of a matching/ranking pipeline with anonymized data and a provided LLM/embeddings API (tools allowed)~30–45 min presentation and discussion of results, ~15 min briefing on the prompt, ~2.5 hours independent work building part of a matching/ranking pipeline with anonymized data and a provided LLM/embeddings API (tools allowed), ~30–45 min presentation and discussion of results, Final stage: Company fit — 30 to 45 minutesValues alignment, collaboration style, and on-site expectations (in person), Values alignment, collaboration style, and on-site expectations (in person)
Day to day: You will design, build, and operate the cloud and data platform that powers Known’s voice-AI product. Most days you will move between Terraform and AWS changes, CI/CD improvements, container orchestration, and monitoring live systems. You will build and maintain data pipelines for ingestion, training data prep, and reporting, then partner with AI/ML to deploy and evaluate models in production. You will sit in the Marina office in San Francisco five days a week, join a short stand-up, align with product on priorities, and close the day by reviewing dashboards and alerts for what you shipped.
Team: You will join a small founding engineering group and report to a founding engineering lead. You will work closely with the CEO and product on priorities, and with peers across backend and AI/ML. Collaboration is tight and hands-on. Everyone ships, reviews code and Terraform, writes lightweight docs and runbooks, and owns measurable outcomes like availability, latency, and data freshness.
Growth: This role has clear paths to Staff or Lead Platform. You can take ownership of the core platform areas: AWS architecture, data pipelines, CI/CD, observability, and security. As we scale, you may lead hiring for a data engineer and help define golden paths that enable other teams to ship quickly and safely. Success looks like reliable SLOs, faster deploys, lower p95 latency, and cost per user that trends down while usage grows.
Ideal Candidate Profile: 3 to 8+ years building production matching, ranking, recommendations, or search in consumer products, Strong Python with PyTorch or TensorFlow and Hugging Face tooling, Hands-on with embeddings, LLMs, and vector search (pgvector, FAISS, Pinecone, or Weaviate), Solid data foundations: feature engineering, labeling/feedback loops, reproducible training, and clear evaluation, Comfortable owning inference services with tight p95 latency and cost targets plus good runbooks and alerts, Able to design offline metrics (AUC, NDCG, MAP) and run A/B tests that tie to real user outcomes, Collaborative with platform/backend to productionize models safely and quickly, RecSys or Search engineer at a social, e-commerce, or dating app who improved ranking quality at scale, Applied ML engineer who built multi-stage retrieval + re-ranking and proved lift online, Conversational AI or voice intake work that improved downstream personalization or matching, Hybrid applied ML + MLOps experience setting evaluation standards and model SLOs
