louis forster

Building the future. A maniacal sense of urgency is my operating principle.

Programming Projects

Treyspace— GraphRAG engine mapping spatial whiteboard elements to semantic nodes. Node.js collaborative backend with Redis locks and version-vector scheme. TypeScript MCP-based tool-execution layer. Deployed on GCP. 26 beta users, 5,000+ AI interactions, 13M tokens processed.View Repo Demo
AI Research Workflow Tool— Full-stack TypeScript React tool that compiles natural language queries into executable workflows (Semantic Scholar search, claim extraction, contradiction checking). LLM generates multi-step pipelines; backend orchestrates tool execution with branching.
NanoGPT Implementation— Trained 50M parameter transformer on A100 compute in PyTorch. Implemented efficient autoregressive sampling with repetition penalties and SFT on UK Hansard corpus for parliamentary syntax generation.

Visual Computing Research Intern - Bristol Vision Institute— Fine-tuned optical flow models (RAFT, DPFlow) on Sintel under synthetic noise augmentation using PyTorch and SLURM-scheduled A100s. Re-engineered recurrent feedback loop for bidirectional warping with flow consistency checks and adaptive occlusion masking. Resolved tensor-shape mismatches in ptlflow. Achieved SOTA on BVI-RLV: 29.22 dB PSNR (+0.13), 0.863 SSIM (+0.010), 0.334 LPIPS (-9.2%).Repo
Teaching Assistant - University of Bristol — TA for 200+ undergraduates across Linear Algebra, C programming, Discrete Mathematics, and Data Science.
Software Engineer Intern - Scribblepad Press — Managed web deployment and site optimisation; increased organic traffic by 119%.

MEng Computer Science with Innovation — University of Bristol, 2022-2026. First Class (Current Avg. 81%). Notable: Data Science (92%), Programming Tools (91%), Machine Learning (87%). Academic Expertise: Computer Vision (multi-view 3D reconstruction via epipolar geometry, custom detection pipelines with 3D Hough accumulator, bespoke NMS in Python/NumPy), HMM-based state inference, Bayesian regression with posterior uncertainty estimation (PyMC).
Balcarras Academy — A Levels 2020-2022. 4 A*s: CS, Maths, Physics, EPQ. Awards: UKMT Silver (x5), Grey Kangaroo Qualifier.

Languages: Python (PyTorch), C/C++, Rust, TypeScript, SQL, Haskell, Triton

ML / Vision: Transformers, Neural Networks, Optical Flow, OpenCV, Bayesian Inference (PyMC), HMMs

Infrastructure: SLURM (HPC), Docker, GCP/Azure, Git, GitHub Actions, Redis, PostgreSQL, Linux