I build AI agents and the infrastructure they run on.
Staff-level software engineer with 20+ years on the web and, since 2023, production AI systems: the first LLM deployment at SecurityScorecard, a RAG report generator before the pattern had a name, and now an agent-ops platform at Toptal that runs coding agents in Kubernetes sandboxes for ~500 clients. Deep real-time browser background (WebRTC, WebSockets), currently applied to multimodal voice agents.
📍 Córdoba, Argentina · remote · open to staff / AI agents / architect roles 🌐 new.picanteverde.com · 𝕏 @picanteverde · ✉️ picanteverde@gmail.com
Agent Ops is DevOps for AI agents: infrastructure that lets executives and engineers run autonomous coding agents on Kubernetes with elastic compute, across model providers.
- Provider-agnostic skills that run the same on Claude Code, Codex, OpenCode and Pi
- Tech lead for multimodal voice-agent research over WebRTC and WebSockets
- Container sandboxing that gives agents full access while keeping the host safe
- A WebSocket chat backed by a Claude Code monitor to steer dozens of agents at once
| 20+ | years building for the web |
| ~500 | clients on the agent-ops platform |
| 700 | engineers trained through the first Toptal Academy React course |
| 50 TB | of network-scanner data processed per day (Scala + Spark) |
| ~85% | reporting spend cut moving Snowflake queries to RDS Postgres |
| Project | What it is | Status |
|---|---|---|
| picante | AI coding agent with an OpenAI-compatible core, skills and built-in tools. Started as an experiment running entirely in Chrome on the File System Access API; now a CLI with a terminal UI. | active |
| picanteverde platform | The site and the services behind it: an auth service with a modules registry, a multi-device app runtime with WebRTC between screens and controllers, and an AI studio that builds apps. | active |
| cv | Resume as data: one resume.json, swappable templates, HTML + PDF output, zero dependencies. |
active |
| Container SSH web terminal | Instantiate containers and reach them over SSH through a browser-based terminal emulator. | planned open source |
| Agent WebSocket chat | Connect to many running AI agents and steer them from one place. | planned open source |
- AI & agents — LLM deployment (vLLM, TGI), RAG, agent orchestration, Claude Code / Codex / OpenCode, MCP servers, LangChain, OpenAI Realtime, Gemini native audio, voice agents, agent sandboxing
- Architecture — enterprise and solution architecture, tech stack evaluation, CI/CD, container strategy, environment isolation, production monitoring, cost optimization
- Backend & cloud — TypeScript, Node.js, Bun, Fastify, GraphQL, Python, AWS, Kubernetes, Docker, Terraform, Datadog
- Real-time & browser — WebRTC (data channels, STUN/TURN, SFU/MCU), WebSockets, WebGPU, WebGL, WebXR, WebAudio, WebAssembly, File System Access API
- Frontend — React, Next.js, Redux, MUI, Svelte, Vue.js, Lit
- Data engineering — Scala, Spark, Airflow, Snowflake, Postgres, ETL / workflow engines, Nearley grammars
Claude Certified Architect – Foundations · Anthropic Education · June 2026 · verify
Architecting Claude-based systems: agentic patterns, tool use, context engineering and production deployment.
Full CV: github.com/picanteverde/cv · I'm @picanteverde everywhere. Friends call me picante 🌶️