Gemini multi-agent runtime
Root ADK orchestrator delegates to corporate, crypto, OSINT, memory, and optional MongoDB MCP specialists.
Google Cloud Rapid Agent Hackathon · MongoDB Track
Auditable trust decisions, not chatbot answers.
Vartovii plans, routes, scores, and stores due-diligence work across company and crypto evidence using Gemini 3.5, Google ADK, Agent Engine, MongoDB Atlas, and MCP.
Live Cloud Run URL
Vertex Agent Engine
5 core agents
28 custom tools
68 tests
MCP MongoDB Atlas
Tech
Gemini + ADK + MCP
Design
Evidence-led console
Impact
Trust decisions at speed
Idea
Cross-domain risk graph
Live Intelligence Overview
Evidence coverage
Current entity universe, trust score distribution, and investigation surface.
—
Companies Monitored
—
Crypto Projects
—
Reviews Analyzed
—
Wallets Tracked
—
Investigations
—
Avg Trust Score
Risk Distribution
coverage pending
CRITICAL: 0
HIGH: 0
MEDIUM: 0
LOW: 0
Submission Evidence
Built to satisfy the judging checklist
Gemini agent runtime, Google ADK deployment path, MongoDB Atlas partner MCP, and verifiable product quality.
Google Cloud deployment path
Cloud Run serves the dashboard and Vertex AI Agent Engine hosts the ADK app object.
MongoDB Atlas MCP
Official mongodb-mcp-server runs through ADK McpToolset when an Atlas connection string is configured.
Quality gate
Automated architecture, service, dashboard fallback, and MCP construction tests keep the demo stable.
Live Source Proof
Live evidence proof, not a static demo
Two live proof paths show current market evidence, real wallet balance evidence, Atlas cache persistence, and the visible agent route behind the decision.
Checking live source
Ethereum
Live price
—
24h movement
—
Market cap
—
Freshness
—
Market signal pending
Source ledger pending
MongoDB persistence pending
Etherscan wallet proof
Checking
Ethereum wallet
Native ETH balance
—
Address
—
Freshness
—
Provider
—
Persistence
—
Agent activity stream
Real-time routing
Trust dimensions
5-axis radar
Agent Workspace
Run the judge demo path
The chat surface is wired to the ADK orchestrator, specialist tools, MongoDB-backed memory, and a session-level audit trail.
AI
Press Enter to send · Shift+Enter for new line
Judge Evidence Trace
One run, full proof path
Compact execution evidence for the judging checklist: route, tools, MCP availability, saved decision, and audit trail.
Trust Leaderboard
Ranked entities
Click any row to inspect the score, risk label, and supporting detail contract.
Auto-refresh in 60s
| # | Entity | Trust Score | Risk Level | Details |
|---|---|---|---|---|
|
Loading leaderboard…
|
||||
Risk Intelligence
Risk distribution
Severity buckets across companies and crypto projects in the active evidence set.
CRITICAL
—
Entities with severe trust deficits requiring immediate review
HIGH
—
Significant risk indicators detected across multiple signals
MEDIUM
—
Moderate concerns — monitoring recommended
LOW
—
Healthy trust scores with strong positive signals
Investigation History
Recent decisions
Saved assessments and summaries produced by the memory agent.
No investigations yet. Use the chat to trigger one!
Multi-Agent Architecture
Agent graph
Domain specialists keep the root orchestrator focused on routing while tool access stays scoped.
Python 3.12
FastAPI
Google ADK 2.0.0
Vertex AI Gemini 3.5
MongoDB Atlas
Cloud Run
MCP Protocol
Docker
Agent Audit Trail
Execution trace
Agent actions, tool calls, model routing, and persistence events.
Standalone Project Context
Standalone hackathon build for evidence-first trust investigations.
This project was built for the contest as a focused Google ADK, Gemini, Agent Engine, MongoDB Atlas, and MongoDB MCP implementation. The domain comes from trust research patterns, but the submitted agent, proof endpoints, UI, tests, deployment scripts, and documentation are packaged as a standalone hackathon project.