I build production AI systems — from RAG pipelines to real hardware.
AI Engineer specializing in LLMs, retrieval-augmented generation, and multi-agent systems, with a rare foundation in hardware R&D and manufacturing automation. I turn research prototypes into reliable, measurable products.
About
I'm an AI Engineer, Machine Learning Engineer, and former TV hardware R&D specialist. My background spans end-to-end ML pipelines, LLM applications, and statistical analysis — alongside the delivery of 35+ TV development projects across Asia and China, from early design to mass production.
In the AI domain I build practical, production-ready solutions: LLM-powered chatbots and multi-agent systems with LangGraph, retrieval-augmented generation with hybrid search, and modular FastAPI backends. My ML work covers NLP, classification, anomaly detection, and deep learning with TensorFlow, scikit-learn, and PyTorch. I also design Python automation that cuts operational workload by up to 70%.
Across every role I focus on data-driven decisions, automation, and quality — turning raw data into insight and integrating AI into real engineering workflows.
Stack & tooling
▹Languages & ML
▹Data science & modeling
▹LLM & agents
▹Retrieval & data
▹Backend
▹Infra & DevOps
▹Observability
▹Engineering automation
▹GTM & data ops
Selected work
Sourcing the World Cup
Directed an AI-agent sourcing pipeline for Startup World Cup across 60+ countries — 490 contacts shipped import-ready, and a data-integrity audit that found only 22% of a 504-row batch had genuine LinkedIn URLs before a single email went out.
Churn Rate Analysis
Built a 90-day churn model for a 1,642-customer SaaS book, restricted to interpretable models only. Found the population was two segments that respond to the same signal in opposite directions — a pooled model would have been wrong for 4,202 accounts — and shipped three routed logistic regressions with a 2.86x lift on the top 150 retention calls.
Autonomous Research Agent
A LangGraph agent that plans, searches the web, and writes structured research reports end to end — a multi-node graph with web search and multi-provider LLMs.
Support Agent with RAG
Customer-support assistant grounded in a knowledge base. Graph compiled once at startup, hybrid retrieval over ingested docs, and streaming chat responses.
Lead-Generation Agent
Automates lead and job discovery with headless browser control, then syncs results to Google Sheets — an agentic pipeline over real-world web data.
Experience
Artificial Intelligence Engineer
- Designed and deployed end-to-end RAG pipelines using LangChain, LangGraph, and Qdrant with hybrid retrieval (dense + BM25 + RRF), integrated into FastAPI services with async PostgreSQL, Celery queues, and Redis.
- Built multi-agent systems with LangGraph state machines — router → tool-executor → responder — including SQL agents, document Q&A tools, and confidence-gated human escalation.
- Owned production backend: SQLite → PostgreSQL migration with Alembic, JWT auth, Docker deployments, and Git Flow versioning for on-premise software.
- Evaluated and configured LLM inference on constrained GPUs (llama.cpp, GGUF/Q4, vLLM) for air-gapped, sovereign deployment.
Growth & GTM Operations
- Directed an AI sourcing agent across 60+ countries in four regions for Startup World Cup, shipping 490 import-ready contacts and 239 events into HubSpot against a validated 17-column schema.
- Ran the data-integrity audit that verified 504 LinkedIn URLs individually and found a 22% pass rate; traced the root cause to synthesized profile slugs and rebuilt the working pool around records that held up.
- Audited 2,364 qualified CRM records against Apollo and identified the 1,138-contact email gap as the binding constraint on outreach scale.
- Deduplicated a 12,415-record sponsorship list with zero loss of same-name-different-data rows, and defined field-validation standards enforced as hard filters rather than post-hoc cleanup.
Process Engineer — Analysis
- Configured and optimized Die, Wire, and Lens Bonding machines, achieving a >95% production yield and improved reliability.
- Verified ISO 9001 compliance and evaluated three SMT PCBA vendors during new-line setup.
- Ran functional testing and performance analysis of optical chip boards, cutting error rates from 5% to 2%.
- Built Excel-macro data acquisition that reduced processing time from 3 hours to 5 minutes.
Research Hardware Engineer
- Led TV development projects across Asia & China — delivered 35 projects (42"–83", UHD & OLED) from design to mass production.
- Specialized in DC-DC IC testing and schematic/PCB design (DFMA); optimized IC selection for 15%+ cost efficiency.
- Ran EMC testing to CISPR32 with >4 dB margin; managed Asia/China regulatory compliance (CCC, RoHS) and secured 4.5–5★ energy certifications.
- Developed Python automation (Paramiko, PyQt, SSH) for AIM/DAJABA projects, enabling 24/7 automated testing with email summaries and Git-based deployment with Korea HQ.
Education & certifications
Education
B.Eng., Electrical & Power Engineering
Have a project or role in mind?
jasman0603@gmail.com