>

ls -la ~/projects

A collection of projects spanning AI/ML systems, low-level programming, and hardware design. Each represents a unique challenge and solution.

$ cat featured.txt

>

Hybrid MCP Hub

Featured

A Local-First AI agent that intelligently routes user tasks between a local model (Ollama/Llama 3.2:1b) and a cloud model (Groq/Llama 3.3 70B) based on privacy constraints using Model Context Protocol.

PythonLangGraphFastMCPOllamaGroqStreamlitMicrosoft PresidioSQLite
>

VAC Protocol: Verifiable Agent Credentials

Featured

Architected a capability-based security framework in Rust that shifts from identity-based to task-scoped credentials, enabling fine-grained policies and instant revocation for AI agents, solving the 'over-privileged agent' problem.

RustAxumTokioBiscuit AuthEd25519WasmtimeDatalogMCPDocker
>

Axiom: Axiomatic Verification

Featured

A protocol-agnostic hybrid verification framework combining symbolic execution (KLEE) and formal modeling (TLA+) for distributed and concurrent systems with bidirectional counterexample exchange.

PythonCTLA+KLEESystemVerilogDockerpytest
>

Garuda: RISC-V INT8 ML Accelerator

Featured

A high-performance RISC-V coprocessor in SystemVerilog that extends RISC-V with custom INT8 multiply-accumulate (MAC) instructions, achieving 7.5-9× latency reduction (p99: 307→34 cycles) for batch-1 attention microkernels compared to standard CPU execution.

SystemVerilogRISC-VCVXIFCVA6VerilatorCocotbPythonCMicrokernel

$ ls other/

>

TraceLens: Visual Debugger for LangGraph Workflows

Created a detailed diagnostic center for LangGraph workflows using OpenTelemetry and SQLite, helping to solve the 'Silent Failure' problem in AI systems by allowing real-time visualization and the ability to go back in time to debug issues.

PythonFastAPINext.jsReact FlowOpenTelemetrySQLiteTypeScript
>

Serverless RAG Chatbot: Scale-to-Zero RAG Architecture

Architected a production-ready RAG chatbot with true scale-to-zero architecture using Next.js 16, Gemini, and Upstash, achieving $0/month cost when idle and sub-50ms cold starts on Vercel Edge runtime, eliminating the $50/month minimum cost floor of alternatives like Pinecone.

Next.jsTypeScriptReactGoogle GeminiUpstash VectorUpstash RedisgRPCVercel Edge
>

BioXAI

A modular framework for explainable AI in biological sequence analysis, integrating protein language models (PLMs), causal inference, and sparse autoencoders for transparent protein function insights.

PythonPyTorchTransformersHugging FaceESM2SHAPLIMEGrad-CAM
>

PSF Design Lab

An interactive toolkit for assessing AI system predictability using the Predictability Spectrum Framework with web app, API server, and browser extension.

TypeScriptReactNode.jsExpressFastAPIGoogle GeminiMaterial UI
>

PDF/DOCX Reader for Cursor

A Python tool to read PDF and DOCX files directly in Cursor IDE, extracting text content with metadata for AI workflows.

PythonpdfplumberPyPDF2python-docxVS Code Marketplace

> More projects available on GitHub

View GitHub Profile