▄ █▄ █▄ █▄ █▄ █▄ ████████████████ ┃━━━━━━━━━━━━┃ ┃ ^ ^ ┃ ┃ === · === ┃ ┃ \____/ ┃ ╲__________╱
Jan Sirui Chen
AI Engineer
Munich, Germany
Building agentic systems at YC and venture-backed startups. Orchestrating coding agents to ship full-stack features. Youngest TA for LLM security at TUM. Previously worked as a crypto analyst at an a16z-backed startup and Baidu.
Experience
AI Engineering Intern
· Comena (YC S25)Feb 2026 — Present
Building full-stack PDF source highlighting with Gemini AI vision for bounding box extraction. Integrated Gemini 3 Flash with coordinate conversion and React-based highlight overlay with scroll-to-view UX.
AI Engineering Intern
· NOA IntelligenceOct 2025 — Feb 2026
Architected event-driven queue systems with PostgreSQL, built autonomous AI agents with LangGraph, and developed LLM evaluation frameworks. Reduced data pipeline complexity by 58% and API calls by 80%.
Teaching Assistant
· TUM — Chair of Software Engineering & AIMar 2025 — Jul 2025
Designed and taught a lecture on LLM Fundamentals for 40 students including Master candidates. Youngest TA (3rd semester BSc) in the chair's history, selected among PhD-level peers.
Product Intern & First Employee
· MoonblockFeb 2023 — Aug 2023
Built and maintained the core database for a proprietary market intelligence tool tracking NFT projects for Fortune 500 clients. Deployed automated data pipelines with Python web scraping and pandas.
Founding Team & Head of Analysis
· UnfilteredApr 2022 — Oct 2022
Shaped the product for a B2B SaaS platform that grew to $240K MRR. Led the core analytical function for 300K+ members, evaluating financial and sentiment data.
Education
BSc Computer Science
· Technical University of MunichOct 2023 — Present
Currently on a gap semester to build AI agents full-time.
TUMKolleg Program
· TUM & Werner-Heisenberg-GymnasiumSep 2021 — Jun 2023
Selective collaborative program between TUM and WHG Garching for exceptionally talented students in the final 2 years of Gymnasium.
Early Enrollment Program
· Technical University of MunichEarly Studies
TUM early enrollment program for excellent high school students in Computer Science. Handpicked students participate in university-level Computer Science courses alongside regular degree students.
Projects
Grand Prize Winner (1st Place overall) · Track Winner
Intelligent freight rate quotation agent. Processes unstructured email requests via GPT-4o-mini extraction, resolves port names through 4-tier fuzzy matching (exact/alias/difflib/LLM), normalizes messy Excel rate sheets (ditto marks, merged cells), and calculates volume-weighted pricing with customer-specific SOP rules. Includes HITL review queue, full audit tracing, and REST API.
#fastapi #streamlit #openai #postgresql #pandas #openpyxl
3rd Place in Databricks Track · Hack-Nation by MIT Sloan · 3000+ participants
Healthcare intelligence platform for the Virtue Foundation. Multi-agent system (OpenAI Agents SDK, GPT-5.2) with debate-driven claim verification (Advocate/Skeptic/Judge), knowledge graph of 742 Ghanaian facilities, medical desert detection, and mission deployment scoring. Full-stack with Next.js frontend, D3 network viz, Mapbox maps, and real-time agent tracing via SSE.
#openai-agents-sdk #nextjs #d3 #networkx #fastapi #lancedB
3rd Place overall · Track Winner · Tech Europe Hackathon
AI property management platform for short-term rentals. Vision AI extracts listing details from live camera walks (OpenCV + multi-provider LLMs via OpenRouter), detects maintenance issues from WhatsApp videos with frame-by-frame analysis, and auto-alerts property managers via Gmail. 5-service Docker Compose architecture with React, FastAPI, Flask, and MongoDB.
#react #fastapi #flask #opencv #openrouter #mongodb #docker
Full-stack Twitter bookmark organizer. Generates semantic embeddings (OpenAI text-embedding-3-large, 1024D), clusters via UMAP dimensionality reduction + HDBSCAN, and auto-names topics with GPT-4o-mini. Interactive D3.js force-simulation bubble map for visual exploration. Built with Next.js, FastAPI, and PostgreSQL with pgvector.
#nextjs #fastapi #d3 #openai #hdbscan #umap #pgvector
Quantum Maze Quest — a 2D maze game with A* enemy pathfinding, dynamic ray-traced lighting (Box2D Lights), timed power-ups, and an integrated level editor with reflection-based serialization. Built with LibGDX/Box2D physics, SpriteCache terrain batching, and a 5-level campaign with contextual audio.
#java #libgdx #box2d #a-star #gradle #github-actions
Recommendation
“Jan possesses a unique blend of deep intrinsic learning, the ability to communicate complex ideas with clarity, and a proactive drive that creates tangible results. He doesn't just complete tasks; he identifies opportunities for impact and builds solutions that elevate everyone around him.”
— Dr. Mark Huasong Meng, Research Fellow, Chair for Software Engineering & AI, TUM
PDF →© 2026 Jan Sirui Chen. All rights reserved.