// ai labs

AI Labs

Building and running AI systems — local-first inference, autonomous agents, and cloud AI tooling.

Local AI Stack

Full offline inference stack. No cloud dependency for day-to-day tasks.

1
2
3
4
5
6

Paperclip Agents (localhost:3100)
  └── OpenCode CLI
        ├── Ollama (localhost:11434) — qwen2.5-coder:7b (GPU, GTX 1660 Ti)
        └── LM Studio (localhost:1234) — gemma-4-12b-qat, qwen3.6-27b (CPU/RAM)

Open WebUI (localhost:13000) — interactive chat, RAG, document Q&A
Service Model Use
Ollama qwen2.5-coder:7b GPU-accelerated coding, refactors, fast inference
LM Studio gemma-4-12b-qat Planning, reasoning, multi-step tasks
LM Studio qwen3.6-27b Complex reasoning (CPU, slow)
LM Studio gemma-4-e4b Fast small tasks
Paperclip CEO + CTO agents Autonomous task execution via opencode_local adapter

Claude Agents

Custom agent configurations built in Claude.ai for real workflows. Each agent has a focused system prompt, the right model, and MCP server connections where needed.

Agent Model Purpose
🔍 Deep Researcher claude-sonnet-4-6 Multi-step web research with source synthesis and citations
📊 Data Analyst claude-sonnet-4-6 Load, explore, and visualize data; build reports from datasets
🚨 Incident Commander claude-opus-4-6 Triage alerts, incident triage, structured response
⚙️ Structured Extractor claude-sonnet-4-6 Parse unstructured text into typed JSON schemas

Claude Code

Using Claude Code CLI as a daily development tool. Every app in Vee Labs — BreachGuard and ACServiceApp — was built with Claude Code handling architecture, refactors, security hardening, and CI/CD setup.

Tooling in use:

Google AI

Certified in Google’s AI agent track. Exploring agentic system design with Gemini and Vertex AI.

Certifications:

Tools & platforms: Gemini · Google AI Studio · Vertex AI · Google Cloud

Prompt Engineering

A working collection of prompts and agent configurations for research, writing, analysis, and decision support — built from real use, not theory.

Approaches explored: