Native Over Wrappers
Heavy AI frameworks hide critical execution paths behind fragile abstractions. I build core orchestration loops natively to retain full visibility over state, latency, and token economics.
Building native AI architectures from first principles.
I'm an AI Systems Engineer based in Indonesia specializing in native multi-agent architectures, resilient backend web services, and data-layer engineering across PostgreSQL, MongoDB, and MySQL.
My core strength lies in architectural design, deep technical research, performance profiling, and security auditing. While I collaborate with frontier AI models as research sparring partners and code accelerators, every architectural decision is engineered from first principles and audited manually line by line before shipping.
Define system topology, database schema contracts, and failure boundaries.
Build lean Python/JS runtimes without heavy black-box orchestration layers.
Manual code review, latency profiling, and red-teaming security checks.
Modern engineering isn't about typing boilerplate faster—it's about knowing what to build, why a design holds under pressure, and where it breaks.
Heavy AI frameworks hide critical execution paths behind fragile abstractions. I build core orchestration loops natively to retain full visibility over state, latency, and token economics.
LLMs are probabilistic; production systems cannot be. I isolate probabilistic reasoning from deterministic runtime validation, atomic DB leasing, and strict security boundaries.
Every system I ship undergoes manual line-by-line code auditing, architectural stress-testing, and empirical validation for performance bottlenecks and security gaps.
A zero-framework multi-agent orchestration engine built from scratch.
Existing agent frameworks often lock engineers into rigid abstractions, opaque retry loops, and unpredictable memory states—making debugging and security governance nearly impossible at scale.
Validates incoming payloads, enforces strict schema budgets, and isolates untrusted prompt context before entering the execution runtime.
Native execution loop supporting three distinct MoA collaboration modes without hidden middleware overhead or opaque retry chains.
Human-readable, inspectable graph-based memory architecture—eliminating black-box vector state drift and enabling Git-diffable state audits.
Built-in verification layers to cross-examine intermediate reasoning, catch hallucinations, and enforce output integrity before downstream execution.
Native MCP (Model Context Protocol) bridging paired with multi-provider BYOK routing so engineering teams never face vendor lock-in.
Autonomous multi-agent OS built on AstraeaADK for solo-founder workflows. Features database-composed runtime roles, atomic Kanban ticket leasing to prevent race conditions, sandboxed execution, and an independent oversight agent.
Local desktop defense engine designed to disrupt classic deepfake and VAE-based manipulation pipelines via ONNX adjoint gradient graphs, DCT Meta-UAP perturbations, and Ed25519 cryptographic provenance seals.
Offline-first, bit-accurate lossless audio player (up to 24-bit/192 kHz) for Windows and Android. Powered by a custom Rust DSP core, 5-band equalizer, and automated FFT spectrum validation to reject fake transcoded FLAC files.
Designing zero-framework orchestration engines, probabilistic turn loops, Mixture-of-Agents topologies, and Chain-of-Verification guardrails from first principles.
End-to-end data modeling, query & index optimization, transactional concurrency control, and hybrid persistent state layers for high-throughput AI and web workloads.
Production mastery in Python and JavaScript for building resilient APIs, asynchronous service pipelines, protocol bridges, and provider-agnostic AI gateways.
Line-by-line manual auditing of complex and AI-accelerated codebases, empirical stress-testing, bottleneck profiling, and strict threat-boundary isolation.
Working proficiency in Rust and Tauri for performance-critical desktop utilities, local ONNX inference graphs, cryptographic tooling, and cross-platform FFI bridges.
Versatile execution across TypeScript, React, Flutter, and PHP to ship interactive dashboards, mobile clients, and legacy backend integrations when projects demand it.
Open for AI Systems Engineering roles, multi-agent & database architecture consulting, and select technical collaborations.