RecallOS Product Recall for AI Decisions

What It Does

RecallOS starts a governed recall case when a faulty AI component is detected. It scans historical decision records, builds Decision Bills of Materials DBOMs, freezes downstream actions, replays decisions under corrected logic, identifies harmed or high-risk outcomes, routes human approvals, performs remediation through UiPath robots and API workflows, validates the fix through Test Cloud, and closes the case with an audit-ready AI Decision Recall Certificate.

Inspiration

Enterprises are moving from AI assistants to AI agents that make and execute operational decisions. But when a faulty AI component is discovered, most tools focus on fixing the live model or monitoring the system. RecallOS focuses on the forgotten question: what about the decisions that already happened?

How We Built It

We built the prototype around UiPath Maestro Case as the control plane. Coded agents perform blast-radius analysis, DBOM generation, deterministic replay, and remediation planning. Agent Builder agents handle intake, impact classification, customer communication drafting, and audit summaries. API Workflows mock LoanCore, CRM, Ticket, EmailDraft, and AuditRegister integrations. A robot simulation updates a legacy portal compliance register. Test Cloud-style checks gate closure.

UiPath Components

• UiPath Maestro Case
• UiPath Maestro BPMN
• UiPath Agent Builder
• UiPath Coded Agents
• UiPath API Workflows
• UiPath RPA Robot
• UiPath Test Cloud
• UiPath Orchestrator
• UiPath Data Service / Data Fabric
• UiPath for Coding Agents

What Makes It Different

RecallOS treats AI incidents like product recalls. It does not let an LLM make financial decisions. Deterministic policy replay produces the corrected outcome, while agents explain, classify, draft, and summarize. The full BPMN model includes Intake, Blast Radius, DBOM, Freeze, Replay, Human Review, Remediation, Test, Certificate, Action Center tasks, and exception loops.

Results

• 120 historical decisions scanned
• 64 decisions used the faulty component
• 11 outcomes changed under corrected replay
• 7 customer-facing high-risk cases
• 44 remediation actions completed
• 25 local verification tests passed
• Successful UiPath Orchestrator job captured for submission evidence

Links

• UiPath Labs Environment: https://staging.uipath.com/hackathon26_244/
• GitHub: https://github.com/xiaodouzi666/RecallOS
• Live demo: https://xiaodouzi666.github.io/RecallOS/control-tower/
• Demo video: https://youtu.be/f9GTKpnuK_I
• Presentation deck: https://github.com/xiaodouzi666/RecallOS/raw/main/deck/RecallOS_Submission_Deck.pptx

What's Next

• Productize every local contract as native Automation Cloud assets
• Replace mock systems with real enterprise connectors
• Add richer DBOM graph exploration
• Expand recall patterns beyond lending to HR, insurance, procurement, and grants

Built With

• codex
• css
• html
• javascript
• python
• uipath-agent-builder
• uipath-api-workflows
• uipath-coded-agents
• uipath-data-service
• uipath-for-coding-agents
• uipath-maestro-bpmn
• uipath-maestro-case
• uipath-orchestrator
• uipath-rpa-robot
• uipath-test-cloud