Experience
Mil-Techs
San Jose, CAAI EngineerJun 2025 – Present
- Built core components of a production natural-language analytics agent in Python/FastAPI that enabled non-technical staff to query 4 internal systems (ERP/MES/PM/WMS) without SQL — 400+ queries/month
- Designed hierarchical schema linking with hybrid BM25/pgvector retrieval and cross-encoder reranking; achieved ~98% gold-column recall within a 3.2k-token schema context budget
- Routed all SQL through MCP tools behind 3 read-only layers and built a log-driven loop regenerating only the failing clause
- Developed a support-ticket triage agent with an LLM, embedding, and rules ensemble across 18 intents; cut first-response time from 1–2 days to seconds and reduced LLM calls from 4 to 1 on the fast path
- Orchestrated 3 specialist agents with asyncio parallel dispatch, A2A calls to the analytics agent, and intent-gated corrective RAG; enforced a rule-based answer-policy guard with confidence-gated human escalation — no incorrect price or lead-time quotes observed across 1,000+ audited tickets
- Shipped both agent systems with 600+ tests in 3-stage GitHub Actions CI, Redis-backed shared circuit breakers, OpenTelemetry/Grafana observability, and an LLM-as-judge regression harness
Blinkle AI
San Jose, CASoftware Engineer InternMay 2024 – May 2025
- Built job aggregation and application-tracking APIs (TypeScript/NestJS/Prisma/AWS) and an LLM resume-parsing pipeline extracting 30+-field profiles from PDF/DOCX via function calling and Zod validation, tested on 500+ real resumes
- Built a Manifest V3 Chrome extension (TypeScript/React) auto-filling applications across 300+ platforms via per-ATS adapters and LLM fallback for custom screening questions
- Tuned PostgreSQL with indexing, partitioning, and materialized views, and layered Redis caching over hot jobs and sessions, which cut query latency by 75% and database load by 40%
FlowMingle
San Jose, CASoftware Engineer InternJan 2024 – Apr 2024
- Built full-stack social features (likes, follows, profiles) across a React/Next.js/TypeScript frontend and NestJS/Node.js backend, shipped 10 REST endpoints on PostgreSQL, and cut feed queries from 40 to 3 by removing N+1 Prisma calls
- Cut latency on location-lookup endpoints 50% by caching Google Maps Geocoding responses in Redis
Kaiyi Electronic Technology
Henan, ChinaSoftware EngineerFeb 2022 – Aug 2023
- Built Python/FastAPI APIs unifying device, firmware, and factory test records across 20+ device models in PostgreSQL
- Implemented Celery/Redis pipelines that processed 5K+ daily production test uploads, validated serial numbers and firmware versions, auto-flagged failed devices for QA review, and generated reports to AWS S3
- Developed React/TypeScript dashboards and cut per-device lookups from 15 minutes of manual search to a single query
Education
Santa Clara University
Santa Clara, CAM.S. in Computer ScienceSep 2023 – Jun 2025
Baekseok University
South KoreaB.E. in Information and Communication TechnologyMar 2018 – Feb 2022
Skills
Languages: Python, TypeScript, JavaScript, Java, Go
AI and LLM: RAG, GraphRAG, MCP, A2A, LangGraph, PydanticAI, vLLM, Langfuse, LLM-as-judge evals, Multi-agent orchestration
Frameworks: React, Next.js, NestJS, Node.js, Prisma, FastAPI, Celery, pytest, Spring Boot, PyTorch
Infrastructure: AWS, Docker, Kubernetes, PostgreSQL, pgvector, Redis, Kafka, GitHub Actions, OpenTelemetry