I build the trust-to-action stack: trusted data → features → evaluated signals → agent actions → accountable human decisions. Live, evaluated systems on Cloud Run + Vertex AI — not demos. Ex-VC / family-office operator, so I also own the layer most engineers skip: getting executives to actually adopt the AI.
📍 Los Angeles, CA · GCP · AWS · Azure · dbt · Vertex AI · RAG · Python
🌐 gozeroshot.dev · 💼 linkedin.com/in/anixlynch · 🦋 anixlynch.bsky.social
| Layer | Question | Live project | Proof |
|---|---|---|---|
| L1 Truth | Can we trust the data? | healthcare-ai-data-engineer | dbt medallion on BigQuery · 55 tests green · quality gate · PII masking |
| L1.25 Context | How should AI agents see it? | ↳ same repo (Feast feature store) | point-in-time-correct patient features |
| L1.5 Signals ★ | What might happen? | healthcare-signal-platform | 5 evaluated signals → agent · anomaly F1 0.85 · cluster silhouette 0.41 (535/40K) · classify ±1-tier 100% |
| L2 Action | What should the agent do? | healthcare-genai-engineer | RAG · BM25+dense+RRF · PII guardrails · CI eval gate |
| L3 Influence | Will humans adopt it? | healthcare-forward-deployed-engineer | VPC deploy · runbook · postmortems · human Approve/Override |
★ The flagship proves it: the live Signal Console runs an ablation — the same Gemini agent decides with the signals vs without. On the ops-capacity case the call visibly flips WATCH → ACT NOW. Signals change the decision; they don't just decorate it. Evals tracked in Weights & Biases, agent calls traced in Langfuse.
Most engineers stop at L2. Most executives start at L3. Few people understand the entire chain — Truth → Context → Signals → Actions → Human adoption. That's the rare part, and it's where 10 years of VC / family-office / CEO-office translation (tech ↔ business ↔ stakeholder) becomes a moat.
AI/Platform: Vertex AI · Gemini · Signal Intelligence · Decision Intelligence · Feature Stores (Feast) · anomaly/cluster/classify/rank GenAI: RAG · hybrid retrieval (BM25/dense/RRF) · agents · tool calling · guardrails · LLM eval · vector search (Chroma/Pinecone/Qdrant) Data: dbt · BigQuery · analytics engineering · medallion · data contracts · governance · Snowflake · DuckDB · Microsoft Fabric · Power BI Cloud/Infra: GCP · Cloud Run · AWS Bedrock · Azure · Docker · FastAPI · GitHub Actions Eval/Obs: Weights & Biases · Langfuse · Ragas-style eval · CI regression gates Languages: Python · SQL
MBA, University of Chicago Booth · JLPT N1 · Authorized to work in the US (Green Card)