Computer Science @ NTOU · Product-minded backend builder · Taiwan
Portfolio · Crypto Flash · Mnemosyne · News Agent · Email
I turn problems I encounter in daily life into reliable, AI-powered tools. I like the part after the demo works: deciding what matters, handling failure, operating the system, and learning from real use.
AI is leverage, not the product. I care about what happens when a provider is slow, a scheduled job is delayed, state must survive another run, or an automated message is not useful enough to keep.
See the systems, real product output, and operating evidence →
Real-time market context for a Telegram group without the raw-feed noise.
Crypto Flash has been running for about two months in a 10-member Telegram group. I use the group as a small, real operating environment: observe how information lands, propose improvements, and refine the delivery format with the people receiving it.
| Ingestion | Runtime | Delivery |
|---|---|---|
| WebSocket reconnects and idle-timeout recovery | Bounded queues and back-pressure control | Telegram throttling and rate-limit retries |
| News, RSS, and YouTube monitoring | Persistent state across scheduled runs | Search, daily digests, and event timelines |
Python asyncio WebSockets Gemini API Telegram Bot API GitHub Actions pytest
Read the case study in the repository →
| Project | Problem | Time horizon | Current evidence |
|---|---|---|---|
| Crypto Flash | Turn high-volume crypto and macro updates into timely context | Minutes | Operating for about two months in a 10-member Telegram group |
| News Agent | Compress software, AI, and startup news into a focused briefing | Daily | Scheduled pipeline with persisted history and Telegram delivery |
| Mnemosyne | Turn words I encounter into structured cards and spaced review | Days to months | Deployed, owner-only alpha that I use myself |
Together, they explore one recurring question: how can a small system turn noisy inputs into useful action at the right time?
I work best when I can help define the problem, challenge assumptions, and own delivery—not only implement a predetermined solution. I prefer small, testable changes; explicit failure boundaries; and claims that match the available evidence.
Backend Python FastAPI REST APIs
Data SQLite Firestore Redis
Delivery Docker GitHub Actions Google Cloud Run Telegram Bot API
Quality pytest unittest Ruff
I am open to early-stage product collaborations, technical co-founder conversations, and backend engineering opportunities with people who care about shipping useful software.
- Portfolio: yili-code.github.io
- Contact: yili.code@gmail.com