Engineer at heart. AI optimist.

I build practical systems for AI agents, personal data, and endurance sports, with a bias toward Rust, deterministic behavior, and efficient use of context.

memory_mcp

Deterministic long-term memory for AI agents.

Stores episodes, entities, and facts as a graph. Tracks both when a fact was true and when the system learned it, while preserving superseded knowledge for historical queries.

My research focuses on reliable agent memory: idempotent ingestion, entity extraction, alias resolution, and context assembly that does not turn every retrieval into another probabilistic guess.

How a fact travels through the system:

flowchart TD
    A[Sources and episodes] --> B[Deterministic ingestion]
    B --> C[Entity extraction]
    C --> D[Entities and aliases]
    C --> E[Bitemporal facts]
    D --> F[Memory graph]
    E --> F
    F --> G[Ranked context]
    G --> H[AI agents]

• rusty-intervals-mcp A compact Intervals.icu interface for AI coaches. Nine high-level intents and deterministic preprocessing turn activity data into concise, decision-ready context.

• rusty_apple_mail_mcp A local, read-only MCP server and CLI for Apple Mail. It reads the local mail index, messages, and attachments without IMAP or AppleScript. Air-gapped.

• rusty_user_voice UserVoice feedback for AI-assisted product workflows, with resumable collection, full idea context, secure tokens, and preview-before-apply changes.

• rusty_azure_devops A compact MCP server and CLI for on-premises Azure DevOps. It supports work item management, search, comments, links, and project lookup through shared business logic.

• tome A minimal Rust CLI for organizing and syncing AI agent skills from GitHub repositories or local files into a central skills directory. It keeps the workflow focused on SKILL.md without marketplaces, format translation, or symlinks.

• Trackly A self-hosted home for GPS tracks. Find trail-running and hiking routes by distance, elevation gain, and slope, then view or share them without a social layer.

• Telemetriq Customizable GPX telemetry overlays for sports videos, rendered with WebGPU and exported in the browser through FFmpeg/WASM.