Building autonomous sub-cortices and custom language runtimes.
Lead systems architect at Tokenectomy Labs. Engineering zero-allocation software repair agents, Tree-sitter AST consensus gates, and multi-tier compilers from Mataram, Lombok.
Architectural Stance & Telemetry
I design autonomous software engineering sub-cortices, compiler runtimes, and verifiable digital forensics platforms under the research banner of Tokenectomy Labs.
My engineering philosophy centers on extreme zero-allocation performance under hard physical limits: deterministic Tree-sitter AST surgery, linear-time grammar scanners, and single-turn token conservation that enables compact models to solve enterprise software tasks with zero dirty diffs.
Every tool is constructed machine-to-machine (M2M) first. Autonomous AI agents operate through deterministic sub-cortices in the background, validating syntax and verifying blast radii before any patch is committed.
Laboratory & Invariants
- Organization
- Tokenectomy Labs (Founder & Lead Architect)
- Academic Affiliation
- Universitas Muhammadiyah Mataram
- Core Invariant
- Zero dirty diffs · Linear AST validation
- Core Languages
- Rust, C, Python, Vidyax (.vx), Shell
- Workstation
- Arch Linux (Rolling) + Hyprland (Wayland)
- Location
- Mataram, Lombok, Indonesia (UTC+8)
Orbital Systems & Runtimes
Research & Publications
Dual Sub-Cortex Architecture for Autonomous Software Engineering: Invariant Mapping, Token Surgery, and Zero-Allocation AST Consensus on SWE-bench Verified
We formulate a zero-allocation Dual Sub-Cortex architecture designed to resolve complex software engineering instances autonomously. By delegating fault localization, context scrub surgery, and Tree-sitter AST validation to an upstream deterministic sub-cortex, the architecture reduces token bloat by over 91% while strictly enforcing zero dirty diff invariants across 500 instances of Princeton SWE-bench Verified.
Deep Space Log — Experiments
Autonomous CLI coding agent equipped with 6 local system tools and strict output formatting.
Floating desktop AI assistant built with CustomTkinter and real-time Groq LPU inference.
Rofi-powered natural language desktop shell translating plain English intents into system calls.
Custom 2D tactical tower defense engine featuring wave balance mathematics in Python.
Aerospace emergency management simulator constructed inside Godot 4.