B.Eng. student in Computer Science and Technology at South China University of Technology (SCUT), Guangzhou (Sep 2023 – Jul 2027). I work on cloud configuration systems, RAG, Machine Unlearning and LLM multi-agent products.
💻 CS student / 🛠️ Backend intern / 🧠 Applied ML (RAG & multi-agent) / Machine Unlearning
- 🎓 South China University of Technology — Computer Science and Technology, undergraduate, Guangzhou
- 🔭 Recently interned at Tencent Rainbow (Qicaishi) on online cloud configuration serving WeChat Pay, Tencent Meeting, WeCom, QQ, and Tencent Cloud
- 🌱 Building RAG and multi-agent systems (GridTrace routing, LangGraph supervisor agents)
- 📫 LinkedIn: chenyang-liao · Email: 202330451061@mail.scut.edu.cn
Stack
Highlight reel. Full CV is on LinkedIn.
Software Engineering Intern · Tencent Rainbow Configuration Center (Qicaishi)
May 2026 – Nov 2026
Rainbow is a Go + tRPC + MySQL + Kafka configuration platform with multi-type configs, auth, multi-protocol delivery, release versioning, and plugin extensions. It handles on the order of 100 million configuration changes per day.
- Built AI Review as a sidecar (main repo / proto / AI module decoupled) so sensitive config changes can be reviewed without touching the serving path.
- Built an onboarding Skill: ramp-up ~2 weeks → under 5 days (~64%); department award (~20% win rate) and reused by the team.
Forward Deployed Engineer (FDE) · BRAVAT (China)
Mar 2026 – May 2026 · AI Efficiency Team
Designed and shipped 0-to-1 AI products for procurement and bidding; 10–20 procurement/business users in steady use, estimated RMB 1M+/year.
- Purchase push-order system: Excel upload, auto-compute, and multi-user live edit in the cloud (replaces offline handoff).
- Intelligent bid-document assembler: upload tenders, parse technical-bid templates via DeepSeek-V4-flash, then assemble components from a shared material library.
Spark Campus Intern · iFLYTEK Marketing
Jun 2025 – May 2026
Owned campus promotion for the iFLYTEK AI Developer Competition.
- Used A/B tests across channels and prioritized high-weight official ones (school WeChat accounts, Youth League), 3× coverage.
- Recruited 412 / 300 participants (37.3% above target).
Dec 2025 – Feb 2026
Rural multi-agent decision system: users design village construction plans under current national policy.
- Hierarchical LangGraph setup: one supervisor agent coordinates eight specialist agents.
- Implemented the Supervisor agent and inter-agent communication.
Curated projects I authored. Role is author / maintainer unless noted.
| Project | What it is | Stack |
|---|---|---|
| OpsWarden | RAG ops assistant (course project ranked 1 / 77): GridTrace two-stage routing (vector-space grids, L1 coarse / L2 fine), then answer or open a ticket | Python, FastAPI, Vue 3, pgvector |
| stock_predictor | Deep-learning stock price forecasts (LSTM) on Tushare data, with visualization for a five-day horizon | Python, LSTM |
| Fund-Performance-Tracking-System | Real-time A-share fund NAV tracking: valuation, holdings, returns charts, Redis-backed | FastAPI, Vue, Redis |
| llm_wiki | Desktop app that incrementally builds a persistent wiki from your documents instead of one-shot RAG | TypeScript |
| AI-Quant-Screener | Local A-share screener and research-report prototype with a multi-agent workflow | Vue 3, FastAPI |
OpenReview: ~Chenyang_Liao3
GridTrace routing used in OpsWarden follows Corrective Retrieval Augmented Generation (arXiv:2402.00751).
