Jiajun (Nick) Huo
I’m an AI engineer with a strong math and systems background, and I enjoy building reliable multi-agent and RAG systems that perform well in production.
About
I’m an AI engineer with a strong math and systems background, and I enjoy building reliable multi-agent and RAG systems that perform well in production. Over the last couple of years, I’ve worked across the loop—from data infrastructure to model-driven agent runtimes—focusing on reliability, observability, and evaluation so improvements are measurable.
I’ve shipped at both venture-backed startups and large enterprises, and I like turning ambiguous requirements into deterministic pipelines: tool/function schemas, robust retrieval, and evidence/refusal gates backed by evals. My recent work includes hardening async agent execution, implementing GraphRAG-style knowledge flows, and building optimization loops that iteratively improve extraction and citation quality.
- 2 years' experience
- Joined October 2026
Industry experience
- Software & Internet
- Computers & Electronics
- Financial Services
- Gaming
- Professional Services
- Education
More experts in Santa Clara, United States
Work experience· 5 roles
Raised agent deployment pass rate from 40% to 85%+ through root-cause analysis, automated builds, and agent runtime hardening. Hardened async execution with timeout controls and error propagation while retaining task results and surfacing failure causes. Automated resource cleanup with persistent job queues, idempotent requests, and retries for recovery from failures. Built agent/runtime observability using OpenTelemetry to expose request traces and agent costs across autoscaled replicas.
Architected a multi-agent system with tool schemas via OpenAI function-calling on a Neo4j-backed closed-loop pipeline. Engineered a DSPy GEPA optimizer to reflectively rewrite extraction prompts, improving edge-direction accuracy from 72% to 86%. Built a GraphRAG pipeline with LLM-extracted typed edges, cycle-detection validation, and prerequisite-chain subgraph retrieval.
Led a 10-person team delivering a production RAG copilot for a global airline from ambiguous requirements. Built a query-rewrite layer mapping user phrasing to domain vocabulary, lifting weak queries above retrieval thresholds. Optimized retrieval with hybrid BM25 + Pinecone vector search and structure-aware chunking, raising top-5 hit rate to 88%. Authored a deterministic locator/evidence/refusal gate using an LLM-judge eval; prompt fixes improved citation accuracy from 60%.
Education
Hire Jiajun (Nick) Huo today
Post a job and invite Jiajun, or contact them directly. No hidden commission.