<?xml version="1.0" encoding="UTF-8"?><rss version="2.0"><channel><title>Claudio Condor — DeFi · AI · Blockchain Engineer</title><description>Technical writing on DeFi engineering, AI agents, ZK proofs, cross-chain systems, and blockchain infrastructure.</description><link>https://condordev.xyz/</link><language>en-us</language><item><title>Pedersen commitments + Groth16 in Cadence: the first real benchmarks</title><link>https://condordev.xyz/blog/pedersen-groth16-cadence-benchmarks/</link><guid isPermaLink="true">https://condordev.xyz/blog/pedersen-groth16-cadence-benchmarks/</guid><description>Zero published benchmarks for ZK in Cadence. I measured them. PedersenBabyJub addCommits ~16 CU, ConfidentialToken transfer ~327k gas, mixer Poseidon cross-VM, and more. What&apos;s feasible, what&apos;s not.</description><pubDate>Wed, 20 May 2026 00:00:00 GMT</pubDate><category>ZK</category><category>cryptography</category><category>Flow</category><category>Cadence</category><category>EVM</category><category>Groth16</category><category>benchmarks</category></item><item><title>Model sweep — what 48 calls across DeepSeek thinking modes taught me</title><link>https://condordev.xyz/blog/model-sweep-deepseek-thinking-modes/</link><guid isPermaLink="true">https://condordev.xyz/blog/model-sweep-deepseek-thinking-modes/</guid><description>Structured sweep across DeepSeek V4 Flash non-think / think-high / think-max on 8 agent tasks. Pass-rate stays high; the real cost is reasoning tokens you pay for and never see. Plus two gateway footguns worth knowing.</description><pubDate>Tue, 19 May 2026 00:00:00 GMT</pubDate><category>LLM</category><category>DeepSeek</category><category>MiniMax</category><category>multi-agent</category><category>cost</category></item><item><title>Verification locks: the pattern that stopped my agents from looping</title><link>https://condordev.xyz/blog/verification-locks-agent-loops/</link><guid isPermaLink="true">https://condordev.xyz/blog/verification-locks-agent-loops/</guid><description>Deep-dive on the single mechanism in FlowCode that did the most to stop runaway iterations. Command-as-fingerprint locks, PIVOT_NEEDED / PROBE_NEEDED / SANDBOX_BLOCKED sentinels, and a bash AST sandbox upgrade.</description><pubDate>Fri, 08 May 2026 00:00:00 GMT</pubDate><category>multi-agent</category><category>LLM</category><category>tooling</category><category>agent design</category></item><item><title>OpenCode fork for Cadence — training small specialized models with multi-agent data</title><link>https://condordev.xyz/blog/opencode-fork-cadence/</link><guid isPermaLink="true">https://condordev.xyz/blog/opencode-fork-cadence/</guid><description>Cadence is a niche language that big LLMs don&apos;t handle well. I forked OpenCode to specialize it — and to collect multi-agent traces as training signal for small domain-specific models. Not released yet, and here&apos;s why.</description><pubDate>Tue, 28 Apr 2026 00:00:00 GMT</pubDate><category>Flow</category><category>Cadence</category><category>LLM</category><category>training data</category><category>tooling</category></item><item><title>Refactoring A0X&apos;s collective brain — from ColBERT to OpenRouter+Cohere</title><link>https://condordev.xyz/blog/colbert-to-openrouter-cohere/</link><guid isPermaLink="true">https://condordev.xyz/blog/colbert-to-openrouter-cohere/</guid><description>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.</description><pubDate>Sun, 08 Mar 2026 00:00:00 GMT</pubDate><category>RAG</category><category>AI</category><category>search</category><category>A0X</category><category>ColBERT</category><category>Cohere</category></item><item><title>Why I&apos;m building privacy on Flow — and why compliance must come with it</title><link>https://condordev.xyz/blog/why-privacy-on-flow-and-compliance/</link><guid isPermaLink="true">https://condordev.xyz/blog/why-privacy-on-flow-and-compliance/</guid><description>Personal motivation for building privacy infrastructure on Flow — Latin American financial context, the compliance reality, why Flow&apos;s cross-VM model is uniquely suited, and what &apos;privacy not impunity&apos; actually means in code.</description><pubDate>Wed, 18 Feb 2026 00:00:00 GMT</pubDate><category>privacy</category><category>Flow</category><category>compliance</category><category>DeFi</category><category>ZK</category></item><item><title>Swarm mode: building teams of agents at A0X</title><link>https://condordev.xyz/blog/swarm-mode-a0x-teams/</link><guid isPermaLink="true">https://condordev.xyz/blog/swarm-mode-a0x-teams/</guid><description>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.</description><pubDate>Sat, 14 Feb 2026 00:00:00 GMT</pubDate><category>AI agents</category><category>multi-agent</category><category>orchestration</category><category>A0X</category></item><item><title>Forking SimpleMem for A0X — extending someone else&apos;s memory system</title><link>https://condordev.xyz/blog/forking-simplemem-for-a0x/</link><guid isPermaLink="true">https://condordev.xyz/blog/forking-simplemem-for-a0x/</guid><description>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.</description><pubDate>Sun, 18 Jan 2026 00:00:00 GMT</pubDate><category>AI agents</category><category>memory</category><category>open source</category><category>LanceDB</category><category>HuggingFace</category></item><item><title>CuyFI: shipping a cross-chain yield agent at ETH Global in 36 hours</title><link>https://condordev.xyz/blog/cuyfi-eth-global-argentina/</link><guid isPermaLink="true">https://condordev.xyz/blog/cuyfi-eth-global-argentina/</guid><description>A postmortem on building CuyFI at ETH Global Argentina — LayerZero bridges, Privy wallets, LangChain agents, and ERC4626 vaults in 36 hours. What worked, what broke, and what I&apos;d do differently.</description><pubDate>Tue, 25 Nov 2025 00:00:00 GMT</pubDate><category>hackathon</category><category>AI agents</category><category>DeFi</category><category>cross-chain</category><category>LayerZero</category><category>Privy</category></item></channel></rss>