Security operations that run without analysts. Document intelligence that never leaves your machine. Memory and tooling for agents that persist. We build software that operates independently.
Core Stack
Production code, not prototypes
Multi-agent security system that handles detection, triage, and response without human intervention. Integrates Wazuh SIEM with LangChain agents. 99.28% accuracy on CICIDS2017 benchmarks.
Desktop RAG application built on Tauri and Rust. Runs Ollama locally for inference—your documents never leave your machine. Multi-query fusion improves retrieval by 35%.
Long-term memory for AI agents via knowledge graphs and local embeddings. Agents remember entities, relationships, and context across sessions. Runs on Neo4j with zero API costs.
Video and image upscaling via Real-ESRGAN, wrapped in a native Rust/Tauri app. GPU-accelerated 4x resolution enhancement with detail synthesis.
Speech-to-text with speaker diarization. Combines Whisper with pyannote.audio to produce transcripts that identify who said what. Runs locally.
MCP server that gives agents access to external tools, APIs, and filesystem operations. Type-safe tool definitions via Zod. Extensible plugin architecture.
Framework for measuring performance disparities in skin cancer detection models across different skin tones. Quantifies bias using ISIC datasets and Fitzpatrick phototype labels.
Multi-agent security system that handles detection, triage, and response without human intervention. Integrates Wazuh SIEM with LangChain agents. 99.28% accuracy on CICIDS2017 benchmarks.
Desktop RAG application built on Tauri and Rust. Runs Ollama locally for inference—your documents never leave your machine. Multi-query fusion improves retrieval by 35%.
Long-term memory for AI agents via knowledge graphs and local embeddings. Agents remember entities, relationships, and context across sessions. Runs on Neo4j with zero API costs.
Video and image upscaling via Real-ESRGAN, wrapped in a native Rust/Tauri app. GPU-accelerated 4x resolution enhancement with detail synthesis.
Speech-to-text with speaker diarization. Combines Whisper with pyannote.audio to produce transcripts that identify who said what. Runs locally.
MCP server that gives agents access to external tools, APIs, and filesystem operations. Type-safe tool definitions via Zod. Extensible plugin architecture.
Framework for measuring performance disparities in skin cancer detection models across different skin tones. Quantifies bias using ISIC datasets and Fitzpatrick phototype labels.
Multi-agent security system that handles detection, triage, and response without human intervention. Integrates Wazuh SIEM with LangChain agents. 99.28% accuracy on CICIDS2017 benchmarks.
Desktop RAG application built on Tauri and Rust. Runs Ollama locally for inference—your documents never leave your machine. Multi-query fusion improves retrieval by 35%.
Long-term memory for AI agents via knowledge graphs and local embeddings. Agents remember entities, relationships, and context across sessions. Runs on Neo4j with zero API costs.
Video and image upscaling via Real-ESRGAN, wrapped in a native Rust/Tauri app. GPU-accelerated 4x resolution enhancement with detail synthesis.
Speech-to-text with speaker diarization. Combines Whisper with pyannote.audio to produce transcripts that identify who said what. Runs locally.
MCP server that gives agents access to external tools, APIs, and filesystem operations. Type-safe tool definitions via Zod. Extensible plugin architecture.
Framework for measuring performance disparities in skin cancer detection models across different skin tones. Quantifies bias using ISIC datasets and Fitzpatrick phototype labels.
Production experience across the modern AI/ML stack—from transformer architectures to deployment infrastructure.