Zhiyuan Li
I'm a computer science student at UC Irvine (class of 2028) working on applied AI: agent memory, LLM evals, and the data pipelines that feed them.
I'm currently an AI engineer intern at Unika Health, where I build patient memory for an LLM system and the pipelines behind it. Before that I was at iHealth Labs working on LLMOps and clinical evals. Outside of work I run my own agents on servers I own, and I'm vice president of Tetris at UCI.
Work
- 2026 – now
Unika Health, AI engineer intern
Built memory consolidation for Context Engine, a patient memory system, cutting daily token use by ~90% from 1B tokens/day. Built an Airflow pipeline that backfilled 112,000+ patient calls across 8 years, a Snowflake-to-MongoDB sync for 30M+ documents, and clinician tooling that has produced 3,000+ ground-truth eval cases.
- 2025
iHealth Labs, software engineer intern
Iterated on patient nudge prompts with clinicians (+8% engagement in A/B tests), rolled out Langfuse tracing across 15+ tools and chatbots, and built the team's first clinical eval set and LLM development SOP.
- 2025 – now
Tetris at UCI, vice president
Grew the club from ~50 to 178 members. I run biweekly events and Swiss-format tournaments.
- 2024 – 2028
UC Irvine, B.S. computer science
Projects
- HyprFeed
Turns tech, AI and robotics news into a narrated daily video briefing. Runs as an Airflow DAG on DeepSeek, self-hosted on a Mac Studio with vLLM and SGLang.
- BaronChairStair
The Tetris at UCI Discord bot and daily puzzle. Re-simulates keypresses through the TETR.IO engine so scores can't be faked; 950+ verified clears, 1,300+ tests, 11 releases.
- DIAYN
Discord Is All You Need. Sweeps 60+ job boards every 15 minutes and DMs friends the roles that match their resume, scored by a self-hosted LLM.
- LCP
Records lectures from the menu bar, transcribes them on a GTX 1070 Ti, and hands them to my agents as context.
- acde
A terminal that knows what my agents are doing. Built on a fork of Tabby.
- Replays
An archive of top TETR.IO league replays, collected to study how the best players stack.