Computer Science undergraduate at the University of Manchester
Building reliable AI systems, full-stack applications, and computer-vision tools
| Project | Focus | Evidence |
|---|---|---|
| Scanly | Computer vision / browser systems | Privacy-first QR decoding pipeline, Web Worker cancellation, cross-browser E2E, and 51/52 internal regression fixtures |
| NoteVault | Full-stack systems | React, FastAPI, Firebase Auth, Firestore, typed OpenAPI contracts, Emulator integration, and production-preview browser tests |
| Hunton Fine-tune | Edge LLMs | QLoRA workflow for structured JSON actions, safety validation, GGUF quantization, and Raspberry Pi deployment packaging |
| RTRL-SnAp Reproduction | Research reproduction | BPTT, exact RTRL, and SnAp-1 on the Copy Task with gradient-consistency checks and experiment logging |
A feature-complete browser QR scanner with a layered heuristic image-processing pipeline. Uploaded images stay local, decoding runs in a dedicated Worker, and the repository includes benchmark, performance, coverage, accessibility, and browser-regression gates.
A production-oriented Markdown notebook with Google sign-in, verified backend identity, user-scoped Firestore access, mutation-safe timeline pagination, bounded search, and a generated FastAPI-to-TypeScript contract.
- Fire External Baseline Reproduction — controlled Direct LLM, BM25, Dense, Hybrid, and E-KELL-style comparison infrastructure with strict experiment, resource-identity, and fail-closed validation contracts. Formal empirical runs remain explicitly pending.
- Aurora Christmas Tree — a Python/Tkinter particle-rendering desktop application with themes, interaction, PNG export, headless preview generation, and CI.
- AI agents with structured routing, retrieval, safety checks, and traceable evidence
- Knowledge-enhanced generation, RAG, GraphRAG, and controlled evaluation
- Real-time recurrent learning and reproducible algorithm studies
- Vision-language models, image-text retrieval, and practical image-processing systems
- Production-oriented software engineering: testing, contracts, security, deployment, and maintenance
Languages
Python · TypeScript · JavaScript · Java · C++ · SQL
AI / ML
PyTorch · QLoRA · LLM agents · RAG / GraphRAG · Computer Vision · Vision-Language Models
Web / Systems
React · Next.js · Vite · FastAPI · Firebase · Firestore · REST / OpenAPI
Engineering
Git · GitHub Actions · Playwright · Vitest · pytest · Linux / WSL · Vercel
- Prefer measurable evidence over broad claims.
- Keep limitations, experiment status, and unsupported conclusions explicit.
- Treat tests, security boundaries, reproducibility, and maintenance as part of the product.
- Email: yangjunjie.lin@student.manchester.ac.uk
- LinkedIn: yangjunjie-lin