AI Infrastructure & Context Engineering

I build local-first middleware to solve the Context Window Bottleneck in LLM workflows. My work focuses on deterministic state management, semantic compression, and offline-capable systems.

I am the maintainer of Vidurai (v2.2.0), a local context ledger for developers.

It acts as a bridge between high-velocity developer telemetry (IDE, Terminal) and limited LLM context windows. Instead of sending raw logs to an AI, Vidurai builds a semantic graph of your work session locally, reducing token overhead by ~70% via the SF-V2 (Strategic Forgetting) algorithm.

Architecture:

• Monorepo Strategy: Unified Python SDK, Daemon, and Typescript Extensions in a single verifiable source.
• Local-First: Zero cloud dependency. Data is stored in ~/.vidurai using SQLite (WAL Mode).
• Protocol: HTTP/JSON-RPC communication between the Python Daemon and VS Code Client.

• Systems: Python 3.9+ (FastAPI, AsyncIO), TypeScript, SQLite
• Infrastructure: GitHub Actions (CI/CD), Docker, PyPI
• Concepts: Semantic Compression, Distributed Topology, Event-Driven Architecture

• vidurai
  • The Monorepo. Contains the Core SDK, Daemon, VS Code Extension, and Browser Connectors.
  • Role: Lead Maintainer
  • Status: v2.2.0 (Gold Master)

Building for a future where AI context is local, private, and infinite.