Writing.
Postmortems, benchmarks, and honest accounts of what I'm building. Mostly ZK + privacy, multi-agent dev, and cross-chain DeFi.
2026 7 posts
- 01 AI / Agents
Model sweep — what 48 calls across DeepSeek thinking modes taught me
Structured sweep across DeepSeek V4 Flash non-think / think-high / think-max modes on 8 agent tasks. Pass-rate stays high; the real cost is reasoning tokens you pay for but never see. Plus two gateway footguns: top-level vs extra_body thinking, and MiniMax x-api-key vs Bearer.
→ - 02 AI / Agents
Verification locks: the pattern that stopped my agents from looping
The single mechanism in FlowCode that did the most to stop runaway agent iterations. The planner-defined verification Command becomes the lock — a fingerprint-matched gate that writers can't bullshit past. Plus PIVOT_NEEDED, PROBE_NEEDED, SANDBOX_BLOCKED sentinels.
→ - 03 AI / Agents
OpenCode fork for Cadence — turning a coding agent into a multi-agent Flow team
Cadence is a niche language. Big LLMs miss the resource semantics consistently. I forked OpenCode, rebuilt the orchestration layer around 5+ specialized roles, and shipped 198 commits in 30 days. Honest about why it's not bundled yet.
→ - 04 AI / Agents
Refactoring A0X's collective brain — from ColBERT to OpenRouter+Cohere
We replaced a self-hosted ColBERT ONNX retrieval pipeline with OpenRouter embeddings + Cohere rerank. The numbers, the reasoning, and honest reflections on production RAG cost vs. performance tradeoffs.
→ - 05 ZK / Privacy
Why I'm building privacy on Flow — and why compliance must come with it
Personal motivation for building privacy infrastructure on Flow — Latin American financial context, the compliance reality, why Flow's cross-VM model is uniquely suited, and what 'privacy not impunity' actually means in code.
→ - 06 AI / Agents
Swarm mode: building teams of agents at A0X
Why multi-agent over single-agent, how we orchestrate parallel specialized agents at A0X, and the lessons from running planner/writer/auditor teams on production tasks.
→ - 07 AI / Agents
Forking SimpleMem for A0X — extending someone else's memory system
What I learned extending an open-source agent memory system and wiring it into a production multi-agent platform. HuggingFace embeddings, multi-tenant LanceDB, and +8.5% F1 over the SimpleMem baseline.
→