IBRAMIND BidBox Agent

Governed tender-evaluation working demo · African Agentic AI Design Challenge · Governance / Bid Box

Human stays accountable.

This public demo uses synthetic tender data only. The agent prepares evidence and risk findings, but never selects or awards a bidder.

Synthetic/open dataMCP toolsAudit logHuman approval gate

1. Synthetic tender package

R1
Bid security
Mandatory

R2
Contractor registration

R3
Quality + HSE method

R4
Similar-project experience

2. Run governed review

The demo executes the same bounded sequence documented in the repository.

R1Bid security must be submitted.MISSING EVIDENCE

R2Valid contractor registration.EVIDENCED

R3Construction methodology addresses quality + HSE.EVIDENCED

R4Similar-project experience.EVIDENCED

Risk summary

{
  "missing_count": 1,
  "mandatory_missing_count": 1,
  "risk_level": "HIGH",
  "unresolved_requirements": ["R1"],
  "decision": "NO_AWARD_DECISION_MADE"
}

Human approval gate

{
  "required": true,
  "approved": false,
  "blocked_action": "FINALIZE_EVALUATION"
}

3. MCP workflow

Human reviewer → agent planner → MCP client → BidBox MCP server → compliance matrix → risk summary → draft evidence pack → human approval.

Tools: load_tender · build_compliance_matrix · summarize_risks · prepare_draft_pack.