$ ls ~/projects --sort=date

< Projects />

Building the future, one commit at a time. Here are some of my open-source projects and research work.

36 total
STATUS:
ALL ACTIVE WIP COMPLETED ARCHIVED
TYPE:
ALL Research (21) Tools (11) Systems (11) AI/ML (11) Blockchain (3)
ACTIVE
AI/ML

DeltaCache

KV cache optimization achieving 10-16x faster warm latency vs vLLM, validated across 84 tests on TinyLlama, Mistral-7B,…

Python vLLM PyTorch Prefix Trie +1
EXPLORE()
ROA-LLM
PUBLISHED ACTIVE
Research

ROA-LLM

A systematic security study formalizing Persistent Semantic Entities (PSEs) as a new attack surface in tool-augmented L…

Rust (PyO3) Python 15 LLMs from 7 vendors Factorial Ablation Design +1
EXPLORE()
EdgeCloud-Reasoning
PUBLISHED ACTIVE
Research

EdgeCloud-Reasoning

Mechanism-designed belief aggregation for edge AI: VCG-based truthful elicitation achieves lower false negative rates t…

Python Mechanism Design DeepSets Transformer
EXPLORE()
PRIVATE ACTIVE
Research

EBM-Market

An energy-based / density-based risk modeling framework that challenges the anomaly=risk assumption in ML: GMM density …

Python scikit-learn (GMM) CatBoost FRED API +2
EXPLORE()
ACTIVE
Research

AgentMesh-STM

The largest empirical study of MARL coordination strategies applied to LLM multi-agent systems: 3,400+ episodes across …

Python OpenAI API Ollama MARL Algorithms (VDN/CommNet/QMIX/COMA) +1
EXPLORE()
Universe-Routing
PUBLISHED ACTIVE
Research

Universe-Routing

A 465M-parameter fine-tuned router that classifies queries into 7 distinct epistemic universes (frequentist, Bayesian, …

Python Qwen-1.5-0.5B HuggingFace Transformers Fisher Information +1
EXPLORE()
PRIVATE ACTIVE
Research

merkle-GPT

Applying cryptographic Merkle trees to LLM weight management: 520-823 MB/s build speed, 0.42% storage overhead, 50-68% …

Python PyTorch SHA-256 SQLite +2
EXPLORE()
ACTIVE
Research

encRec

Trie-routed recommendation under encrypted features for privacy-preserving ML

Python PyTorch DeepFM DLRM +1
EXPLORE()
PRIVATE ACTIVE
Research

LinkerLLM

Multi-model LLM serving system with weight section sharing inspired by runtime linkers

Python PyTorch CUDA Transformers
EXPLORE()