Mythos Data Intelligence Agent

DataHub Agent Hackathon 2026 — Challenge: Agents That Do Real Work

An autonomous AI agent that enriches DataHub metadata with real-world context from on-chain data, web sources, and knowledge bases — then writes findings back to DataHub as descriptions, tags, and glossary terms.

OK DataHub MCP Server
OK Claude Agent SDK
OK minia2a x402 Services
OK Open Source (Apache 2.0)

Scenario 1 — Missing Metadata Audit

Finds datasets lacking descriptions, owners, tags, or domain assignments

=== SCENARIO 1: Missing Metadata Audit ===

Found 4 domains:
  Revenue & Finance (42 entities): Revenue analytics, transactions
  Marketing (18 entities): Campaign performance, attribution
  Product & Engineering (31 entities): Telemetry, deployment logs
  Customer Success (12 entities): no description

20 total datasets discovered
  fct.transactions — complete
  dim.customers — missing: description, tags
  campaign.attribution — missing: description, owner, domain
  product.telemetry — missing: tags
  revenue.monthly — complete
  ...

Result: 12 entities need enrichment

Scenario 2 — Auto-Enrichment

Generates suggested descriptions and tags from external context

=== SCENARIO 2: Auto-Enrichment ===

Enriching: dim.customers
   Suggested description: "Customer dimension with demographic
   data, segmentation attributes, and lifetime value scores.
   SCD Type 2 for tracking changes over time."
   Suggested tags: [customers, dimension, scd2, crm, gdpr]
   Suggested domain: Revenue & Finance

Enriching: campaign.attribution
   Suggested description: "Marketing campaign performance metrics:
   impressions, clicks, conversions, ROAS. Joined with ad platform
   data for multi-touch attribution."
   Suggested tags: [marketing, campaigns, attribution, roas, ads]
   Suggested domain: Marketing

Result: 5 datasets enriched with suggestions

Scenario 3 — Lineage Impact Analysis

Traces upstream/downstream dependencies and flags breaking changes

=== SCENARIO 3: Lineage Impact Analysis ===

Root entity: fct.transactions

   Upstream dependencies: 3
      staging.raw_events
      staging.cleaned_orders
      dim.products
   Downstream dependents: 5
      revenue.monthly
      revenue.dashboard
      fraud.detection_model
      marketing.roi_report
      exec.kpi_summary

   Impact: HIGH — 5 downstream entities affected by changes

Architecture

+-------------+     +------------------+     +-------------+
|  DataHub    |---->|  Mythos Agent    |---->|  External    |
|  MCP Server |     |  (Claude SDK)    |     |  Sources     |
|  (metadata) |     |                  |     |  (RPC/API)   |
+-------------+     +--------+---------+     +-------------+
                             |
                    +--------+---------+
                    |  DataHub REST    |
                    |  (write-back)    |
                    +------------------+

Quick Start

# 1. Start DataHub locally
pip install acryl-datahub && datahub docker quickstart

# 2. Clone and install
git clone https://mythos-datahub-2026.s3.us-east-1.amazonaws.com/repo.git
cd repo.git && npm install

# 3. Run enrichment agent
node agent.js --enrich --domain "Revenue & Finance"

# 4. Run full demo
node agent.js --demo

Resources

Built by Mythos Agent · minia2a.uk · DataHub Agent Hackathon 2026