AI parsing accuracy 95%+, bookkeeping 2h→30min; expanded from a single tool into a complete three-role management mini-program
Stack: WeChat Mini Program + Cloud Development + DeepSeek; frontend / backend / database / deployment all built independently full-stack.
Key implementation: three-role permissions, greenhouse visualization, seedling-age back-calculation; AI bookkeeping/order recognition uses a three-layer "rule pre-processing + segmentation + AI fallback", with error feedback (corrections) making the model more accurate over time; shipping/relocation is fully idempotent to prevent duplicates.
Engineering trade-offs: individual account cannot use getPhoneNumber → manual role selection; built order scheduling and auto-reminders for the real scenario of "11 greenhouses, 264 beds".
Keywords WeChat Mini Program · Cloud Development · DeepSeek · Idempotency Design · Three-layer Parsing
Horizon 72h Hackathon | OPC Founder AI WorkbenchBackend Lead2026.07
Delivered backend 0→1 with 15 APIs in ~24 hours
Used Claude Code as the primary coding agent to run "AI generate → human review → hand-written tests → bug localization → fix", with hand-written tests catching 3 key edge cases including a false req.on('close') trigger.
Wrote HANDOFF.md to structure context so a new teammate / agent can take over in 3 minutes.
Ran the full SO101 robotic arm ACT model inference & control pipeline, and rewrote the debugging docs.
Smart Agriculture Research Institute | Data Annotation Intern2026.01 – 2026.02
Completed ~4000 annotation points/day at 95%+ accuracy, understanding AI model quality from the data side.
Education
Tianjin Normal UniversitySoftware Engineering | Bachelor2024 – 2028
Academic Performance: Year 1 GPA rank 16/50, comprehensive rank 7/50; Year 2 GPA rank 13/50, comprehensive rank 9/50 (top 20% overall); school scholarship for two consecutive years