Samaritan is a production-grade multi-agent AI system for OSINT investigations, due diligence, and intelligence research.

Samaritan runs on a self-hosted OpenClaw deployment on top of Parrot OS and orchestrates a fleet of 9 specialized AI agents, each isolated in its own workspace with defined roles, behavioral guardrails, and execution contracts.

A central orchestrator (Seer) routes and delegates tasks to domain specialists covering identity lookup, skip tracing, competitive intelligence, opposition research, pre-employment screening, romance scam detection and more.

• Architected the full agent fleet, workspace isolation, prompt engineering for each specialist role, delegation contracts, and routing logic across all 9 agents
• Engineered a court-defensible case hygiene protocol with automated session resets, SHA-256 signed evidence packages, and per-case audit manifests
• Integrated local LLM inference (LM Studio + Qwen3 / GPT-5.3) with Nomic embedding-based semantic memory for long-term agent context
• Deployed and branded a self-hosted Mission Control dashboard for real-time fleet monitoring, session visibility, and agent chat,  zero cloud telemetry
• Diagnosed a fleet-wide mid-execution failure caused by a runtime workspace inheritance bug, submitted a comprehensive bug report, and applied a targeted bundle patch without destabilizing the system
• Built a production upgrade runbook for a 9-agent AI fleet with automated verification gates, rollback-on-failure, and signed preflight bundles, zero manual recovery steps required on failure