Juan Pablo Enriquez@jpablortiz96
1
Events attended
1
Submissions made
4+ years of experience
About me
I’m a developer passionate about building practical solutions with AI, software engineering, and scalable technologies. I enjoy turning ideas into working products, collaborating with teams, and learning fast in hackathon environments. I’m especially interested in creating impactful tools that solve real-world problems.
🤓 Latest Submissions
ROCm AgentOps
ROCm AgentOps Command Center is an operational assurance layer for AI agents running on AMD-backed inference infrastructure. Instead of treating an agent as a black box, the system scores incidents deterministically, detects risk flags, calculates trust and confidence, routes work to the right execution path, and generates audit-ready outputs. The platform ingests business incidents, live AMD/vLLM workload signals, endpoint health checks, benchmark evidence, and optional ROCm/GPU telemetry. It then compiles each workflow into a safe execution plan: deterministic rules for low-risk cases, smaller-model summaries where speed matters, Qwen 7B through an OpenAI-compatible vLLM endpoint for higher-risk critique and narrative, and human review for critical or low-trust incidents. The Command Center includes a Model Router, What-if Strategy Simulator, Agentic SLA Monitor, Policy-as-Code guardrails, owner-aware escalation packets, SHA-256 audit seals, War Room Packet export, and a Build-in-Public telemetry card. In our validated AMD/vLLM run, Qwen/Qwen2.5-7B-Instruct achieved 20/20 successful benchmark requests, with live incidents generated from benchmark p95 latency evidence. The goal is to make AI agent workflows more trustworthy, auditable, cost-aware, and operationally useful for real teams—not just impressive demos.
10 May 2026
