LLM / Agent Application Engineer
9+ years in NLP · now building LLM & Agent applications for enterprise
NLP → LLM → RAG → Agent → Enterprise Agent → AI × Business
Python · FastAPI · Vue 3 · TypeScript · Deterministic SQL · Human-in-the-loop
Turns a natural-language risk-metric requirement into a schema-validated Metric IR, deterministic SQL, automated tests, statistical experiments, evidence-driven reflection and governed, versioned metric releases.
LLMs reason. Python verifies. Humans govern.
116 API routes · 675 backend tests · content-addressed approvals · full audit trail · runs fully local on synthetic data (no API key, no cluster required).
NestJS 11 · TypeScript · LangGraph · ReAct · MySQL · Redis
Autonomous ReAct agent for enterprise data analysis, risk assessment and financial consulting. 20+ tools with Zod-validated schemas, multi-turn session management and real-time SSE streaming — no predefined intent classification.
Built as an explicit LangGraph StateGraph (model ⇄ tools) rather than a
black-box wrapper, with a deterministic output sanitizer that rewrites internal tool /
field names into business terms and is safe under token-level streaming, plus a
circuit breaker (≤ 15 tool calls, ≤ 3 per tool) against runaway loops. Public version
is fully desensitized.
Python · LangGraph · RAG · ChromaDB · Qwen · Middleware
Customer-service agent for robot vacuums: ReAct loop + RAG over a product knowledge base, 7 tools, and a middleware layer handling tool-call monitoring, dynamic prompt switching and log redaction.
FastAPI · React 18 · LangChain · ChromaDB · JWT
Multi-user / multi-session RAG Q&A system with document ingestion (PDF / DOCX / CSV / XLSX), SSE streaming answers, citation tracing and role-based knowledge-base management.
LLM Applications · AI Agents · ReAct · LangChain · LangGraph · Tool Calling ·
RAG · Vector Databases · Prompt Engineering · Agent Middleware · Agent Runtime ·
Human-in-the-loop · Evaluation & Guardrails
Python · TypeScript · FastAPI · NestJS · Vue 3 · React · MySQL · Redis ·
SQLite · Docker · GitHub Actions · Pydantic · Zod
GitHub-visible timeline of a 9+ year journey.
2018 ─ 2019 NLP keyword & topic extraction · summarization ·
multi-label classification · image labeling (157 commits)
2025 ─ LLM + Web Selenium-driven LLM web automation & tool use
2026 ─ RAG enterprise knowledge-base Q&A with citations
2026 ─ Agents ReAct agents in both Python and TypeScript
2026 ─ Enterprise AI business advisory agent ·
risk-metric engineering platform
I didn't learn LangChain to chase a trend — I got here from years of NLP, one layer at a time.
Open to LLM / Agent Application Engineer roles.