Louis Forster

Louis Forster

Applied ML and systems work. Recent projects include graph-based retrieval for collaborative AI software and offline local-LLM tooling for Jetson devices.

EmailGitHubLinkedInResume

Projects

  • OpenJet

    Link

    Offline terminal-based agent for self-hosted LLMs. Supports Unified memory systems (Nvidia Jetson), Cuda, ROCm, and Vulkan backends.

    • - Decode speed of 39 tok/s for Qwen3.5-27B on RTX 3090.
    • - Automated local model setup via llama.cpp with hardware profiling for optimised configurations
    • - Memory management to avoid OOM errors: Automatic model offloading, context size calculations, automatic context compression
    • - Designed with edge devices and IoT in mind: user can register IO hardware, OpenJet logs the output, and the local LLM reads the log and runs appropiate tools.
    • - Supports agent tool execution, local context management, and approval gates for state-changing actions
    • - Python SDK exposes hardware profiling, background agent orchestration, and model tok/s benchmark parameter sweeps

  • Treyspace

    Link

    AI-Native canvas. Turns Excalidraw canvases into a queryable knowledge graph with semantic, relational, and spatial retrieval.

    • - Over the course of 10 weeks, Built a canvas-to-graph pipeline for Excalidraw JSON data which adds custom fields and clusters suitable for retrieval. Implemen ted GraphRAG to ranks graph clusters relevant to user query.
    • - Production deployment supporting concurrent collaborative sessions and real-time data syncing on Azure/Google Cloud.
    • - Solves the problem of switching between LLM tab and whiteboard tab by integrating it in whilst retaining the crucial connections and spatial data that undermines the visual components.
    • - Helps keep the whole canvas in context which creates connections that the user might not see.
    • - Retrieval quality improved by combining semantic similarity with links and spatial grouping

Experience

  • Visual Computing Research Intern

    Bristol Vision Institute

    Jun-Jul 2025

    Fine-tuned RAFT/DPFlow optical flow models using Sintel dataset with synthetic noise. Implemented bidirectional warping and forward-backward consistency checks and adaptive occlusion masking into the main denoising model. This stabilised the self-supervised training pipeline, surpassing the prior state-of-the-art. Work is a key contribution to a IEEE IMCE 2026-accepted paper (as second author).

  • Teaching Assistant

    University of Bristol

    Sep 2023-Apr 2025

    Teaching Assistant for up to 200 undergraduates across Linear Algebra, C, Discrete Mathematics, and Data Science

  • Software Engineer Intern

    Scribblepad Press

    Jul-Sep 2024

    Managed web deployment and site optimisation. Increased organic traffic by 119%.

Contact: louisforster64@gmail.com