Official portfolio and systems telemetry interface of M N DAFFA (NaDev) — Lead Systems Architect at Tokenectomy Labs, compiler engineer, and autonomous AI researcher.
🌐 Live Interface: daffa2555.github.io
Zero-Allocation Dual Sub-Cortex Autonomous Software Engineering Agent evaluated on 500 instances of Princeton SWE-bench Verified.
- Research Publication: Zenodo DOI: 10.5281/zenodo.22929676
- Model Checkpoints: Hugging Face (NadevA23/Kronumos-Kairos-v2)
Extreme zero-allocation M2M Model Context Protocol (MCP) server suite delivering Tree-sitter AST syntax healing and linear-time token surgery for autonomous coding agents.
Multi-tier programming language constructed from scratch with 4 execution engines (Tree-walker, Python transpiler, C bytecode VM, and native compiler), featuring slot-based escape analysis and autonomous agent execution blocks.
- Documentation: vidyax-lang.github.io/Vidyax
Cryptographically verifiable Domain-Specific Language (DSL) and interpreter for auditable digital forensics investigations with automated chain-of-custody logging.
Specialized coding intelligence model fine-tuned from Qwen2.5-Coder-7B via 4-bit distributed QLoRA across restricted GPU clusters, deployed as real-time Continue IDE backend via vLLM.
- Languages: Rust, C, Python, Vidyax (
.vx), Shell, JavaScript (WebGL/Three.js) - Architectures: Bytecode Virtual Machines, Tree-sitter AST Grammars, JSON-RPC 2.0 (MCP Protocol), Distributed LoRA / QLoRA
- Environment: Arch Linux (Rolling) + Hyprland (Wayland), Neovim, Docker
- Invariants: Zero dirty diffs, linear-time AST validation, sub-millisecond execution latency
Designed and engineered by M N DAFFA (NaDev). Built with Three.js WebGL and vanilla modern standards.