Pinned Loading

1. Interpreting-LoRA-Fine-Tuning Interpreting-LoRA-Fine-Tuning Public

   Interpretability study of LoRA fine-tuning in LLMs: where LoRA puts its capacity (per-module update norm + effective rank) and how much it shifts representations (layer-wise drift), with rank/modul…

   Jupyter Notebook 1

2. LLM-Agent-Toolbox LLM-Agent-Toolbox Public

   Building blocks for LLM multi-agent systems: semantic tool selection (embeddings + recall@k) and a blackboard coordination primitive. From scratch, tested.

   Python 1

3. Medical-RAG-Knowledge-Graphs Medical-RAG-Knowledge-Graphs Public

   Multimodal medical RAG: ColPali visual retrieval + Google Gemini + a Mondo Knowledge Graph, evaluated on the MIRAGE benchmark.

   Python 2

4. Bayesian-Learning-of-Neural-Interactions Bayesian-Learning-of-Neural-Interactions Public

   Bayesian filtering that recovers the interaction structure of a network of neurons from their spike times: exact posterior, consistency with a CLT rate, and an SMC extension. Paper + reproducible c…

   TeX

5. GPT-from-Scratch GPT-from-Scratch Public

   A modern decoder-only Transformer (RoPE, RMSNorm, SwiGLU, KV-cache) implemented and trained from scratch in PyTorch on TinyStories, with a copy-task smoke test.

   Python

6. video-modality-diagnostics video-modality-diagnostics Public

   Do video models actually use the video? A modality-ablation, contribution and robustness diagnostic for multimodal VideoQA (vision/audio/subtitles). Offline-testable + Colab VLM backend.

   Jupyter Notebook