Automation, agents, tooling.
Technical builds and experimental software. Agent frameworks, workflow automation, applied model integration, and structures designed to solve real problems instead of generating noise.
Selected technical work.
Each build started with a real problem and shipped as a real system. No prototypes passed off as products.
Multi-Agent Orchestration Framework
A modular agent framework for coordinating multiple LLM-powered agents across tasks, with built-in memory, tool use, and structured handoff between specialized agents. Designed for consulting workflows that need more than a single-prompt chain.
Voice-Driven AI Assistant Pipeline
End-to-end voice chatbot system connecting speech-to-text, LLM reasoning, and text-to-speech with low-latency streaming. Built for real-time conversational interaction with custom voice models and tool-calling capabilities.
Automated Workflow Engine
Business process automation system integrating calendar, email, CRM, and task management APIs into a unified decision layer. Handles scheduling, follow-ups, data sync, and escalation without manual intervention.
Drone Telemetry & Calibration System
Custom firmware interface and telemetry dashboard for DJI drone fleet management. Real-time sensor data visualization, automated pre-flight calibration checks, and flight log analysis.
Local LLM Deployment & Fine-Tuning Kit
Toolkit for deploying and fine-tuning open-source language models locally. Includes quantization pipelines, evaluation benchmarks, and a clean inference API for running models offline without cloud dependency.
Every build follows the same principles.
Start with the real problem, not the tool. Understand what the system actually needs to do before choosing the stack. Most automation fails because people build for capability rather than for the specific workflow that needs to work.
Ship modular. Every component should be replaceable, testable independently, and designed to extend. No monoliths, no vendor lock-in, no hidden dependencies that collapse when one service changes its API.
Document everything. If someone else cannot pick up the build and understand it in under an hour, the build is not finished.