dipam paul

KVSplit. Run larger LLMs with longer contexts on Apple Silicon by using differentiated precision for KV cache quantization. KVSplit enables 8-bit keys & 4-bit values, reducing memory by 59% with <1% quality loss. Includes benchmarking, visualization, and one-command setup. Optimized for M1/M2/M3 Macs with Metal support.

AgentGuard. Real-time guardrail that shows token spend & kills runaway LLM/agent loops.

synapz. synapz is a research prototype exploring how large language models can adapt teaching content to different cognitive styles. built over a 48-hour sprint with a strict $50 api budget, this project implements a scientific framework to test whether adaptive teaching produces measurably better results than static approaches.

democlean. Score robot demonstrations by motion quality. Identify jerky, inconsistent episodes before training.

cgcp. Enterprise governance control plane for Claude AI. Transforms Anthropic's Responsible Scaling Policy into enforceable controls. Features: real-time risk detection, tier-based thresholds (general/enterprise/research), human-in-the-loop escalation, automated compliance reporting. Deploy in 30 seconds.

k8s-cleanroom. Python

chefscale. Swift

lie-circuit-mats-2025. repo for MATS 2025

morpho-images. Jupyter Notebook