Azan Hyder. I turn research questions into working software, with a focus on systems that are measurable, inspectable, and useful beyond the demo.
Metaorcha — many models, one harness. Building Orcha, an Apache 2.0 runtime where one goal becomes a verified multi-agent run across MCP, A2A, and Computer-Use.
| Stage | Milestone |
|---|---|
| 🟢 Now | v1 runtime: multiprotocol orchestration, shipping locally |
| 🟡 Next | v1.2 harness: DAG execution, output verification, retry and fallback |
| 🔵 Aim | network layer: peer discovery, fulfillment, reputation |
|
🤖 AI and agent systems Agent orchestration, MCP tool ecosystems, retrieval, and verification loops for more dependable model behavior. |
🔗 Decentralized systems Experiments in governance, trust-minimized exchange, zero-knowledge proofs, and fairer economic coordination. |
📈 Quantitative research tools Probabilistic models, backtesting systems, and analysis software that make assumptions visible. |
The Harness Layer on dev.to: research field notes on agent harnesses, each with sources, a limitations section and a reproducible artifact. Latest: When should an agent stop? (artifact in code-desk-cli/research). Practice pieces and short takes on X.
|
Code Desks Run an office of AI coding sessions from your terminal: two hats, gated handoffs, and a meter that reads what each desk really used. Python Agents MIT |
Metaorcha · metaorcha.ai Apache 2.0 runtime for multi-protocol agent orchestration. Python Agents |
|
CDV A frozen public research preview for judging AI-agent work with a deterministic floor, an LLM critic, and Bayesian stopping. Python LLM evaluation |
ML Trend Probability A machine-learning signal framework that tunes and stacks trading indicators into a trend-probability strategy. Python ML Quant |
|
SDCA RAQQR A Bitcoin market-risk model and backtesting library for accumulation and distribution strategies. Python Quant research |
If you are working on agent runtimes, protocol bridges, or verification, open an issue on Metaorcha or send me an email.



