Skip to content

Projects

Everything I have built or am still building, searchable by node, tag and stack.

14 of 14 projects

  • 2026

    AI Mathematical Olympiad: Progress Prize 3

    Getting an LLM to solve olympiad-level maths inside an offline Kaggle kernel. A QLoRA fine-tune of Qwen3-30B-A3B, vLLM inference with tool-integrated reasoning, and majority voting.

    Qwen3-30B-A3BUnslothQLoRAvLLMSource

  • 2026

    Customer Analytics with Deep Learning

    Segmenting UK e-commerce customers into behavioural profiles from what they buy. MiniLM embeddings of item descriptions, K-Means, and a self-organising map on the way.

    MiniLMK-MeansSilhouette scoringSelf-organising mapsSource

  • 2026

    Gaby AI: the data-cleaning gatekeeper

    Tom's database gatekeeper. A self-maintained data workflow plus a small ranking model inspired by PageRank. Born from asking why the boring Monday-morning pipeline fixes still aren't automated.

    BigQuerySelf-attentionPCAAgentswhoamimi/databy-ai-gaby

  • 2026

    Kaggriculture: the Kaggle farmer

    An agent for a two-player farming sim that has to finish a 30-day season with the most money. Right now it plants wheat everywhere. Heuristics first, then bandits, and RL last.

    PythonHeuristic agentsMulti-armed banditsSource

  • 2026

    Kobi: AI Indexed Ready Docs

    Shared knowledge base between user and AI web/desktop chat provider e.g. ChatGPT, Claude.

    VitepressMarkdownMkDocs-materialFull-stackShadinUIwhoamimi/kobi

  • 2026

    March Machine Learning Mania 2026

    Predicting the win probability of every possible NCAA tournament matchup. A logistic map from Massey rank difference, with bookmaker odds as a sanity check.

    Logistic regressionMassey ordinalsBookmaker oddsSource

  • 2025

    Brain-to-Text '25

    Decoding attempted speech from brain-implant neural signals into text. A pretrained RNN does the decoding, then an online n-gram model corrects what it gets wrong.

    PyTorchRNNN-gram LMEnsemblingSource

  • 2025

    Hedge Fund Time Series Forecasting

    Forecasting a financial target per entity ID across several time horizons with a LightGBM pipeline. Next up: narrow to horizon 1 and add lag features.

    LightGBMFeature processorsPandas

  • 2025

    Helios Corn Climate Challenge

    Building climate-risk features that move with corn futures. F-score selection, correlation-weighted signals, and a homemade copy of the competition metric so I can validate offline.

    Pandasscikit-learnF-score selectionCorrelation metricsSource

  • 2025

    Adaptive Immune Profiling Challenge

    Classifying immune receptor repertoires from V/J gene usage and CDR3 sequences. Still at the EDA stage, with an n-gram kernel half-built.

    PyTorchSequence embeddingsN-gram kernelSource

  • 2025

    MAP: Charting Student Math Misunderstandings

    Reading a student's written maths explanation and deciding whether it's correct, a slip, a named misconception, or none of those. Two stages: correctness first, then the misconception.

    Sentence embeddingsPhi instructTwo-stage pipelineSource

  • 2025

    RSNA Intracranial Aneurysm Detection

    Detecting intracranial aneurysms from CT/MR brain scans. So far: DICOMs into 3D volumes, and zero-shot BiomedCLIP scoring of slices. The detection head doesn't exist yet.

    pydicomBiomedCLIP3D volumesSource

  • 2025

    TaroTarot AI: Fortune Teller

    Reading cards the way I do, pattern inside randomness, then asking an LLM to do the same. My predictions come true 99% of the time according to friends and strangers, so now I'm testing whether a model can.

    Next.jsFastAPIOllamaLightning StudioSource

  • 2023

    Tom: an AI that's meant to become someone

    The main objective to the microservices and projects that I build in this lifetime.

    LLMsMemory networksReinforcement learningPythonSource