Meshia Data Recovery Agent

Evidence-first incident recovery using DataHub context, lineage, bounded action, independent verification, and trusted write-back.

Comment

Meshia Data Recovery Agent

Meshia Data Recovery Agent is a new, isolated hackathon project for the Agents That Do Real Work track.

It uses DataHub as the operational context layer for data incidents. Before proposing any action, the agent collects schemas, lineage, ownership, domains, documentation, and quality signals. It separates verified facts from inferences, calculates the downstream blast radius, produces a bounded remediation plan, verifies the outcome independently, and records trusted incident knowledge back into DataHub.

Problem

Data incidents are often handled with incomplete context. Operators must manually discover which assets are affected, who owns them, what depends on them, and whether a proposed fix actually solved the problem. AI agents without reliable metadata can hallucinate dependencies or act beyond their authority.

Solution

The project turns DataHub's metadata graph and agent tooling into an evidence-first recovery workflow:

  1. Receive a data incident or schema-change signal.
  2. Search DataHub for the affected asset.
  3. Collect schema, ownership, domain, documentation, quality, and lineage evidence.
  4. Trace downstream dependencies and calculate blast radius.
  5. Separate facts, inferences, unknowns, and blockers.
  6. Generate a bounded remediation plan with no uncontrolled production authority.
  7. Run independent verification checks.
  8. Produce a final PASS, WARN, FAIL, or UNVERIFIED status.
  9. Write the verified incident outcome back to DataHub as durable context for future humans and agents.
  10. Create a deterministic evidence seal for the complete run.

Planned DataHub use

  • DataHub OSS / Core Platform
  • DataHub MCP Server
  • Search and entity inspection
  • Schema and ownership context
  • Multi-hop lineage and impact analysis
  • Governed write-back or proposals
  • Saved incident knowledge and evidence references

Safety boundaries

  • The private Meshia production core is not included or accessed.
  • The public repository contains no secrets, personal data, private logs, or backups.
  • Production mutation authority is denied by default.
  • Write operations require an explicit approval gate.
  • Mock and sandbox modes are available for safe judging and testing.
  • Any reused concepts or pre-existing framework code will be disclosed transparently.

Target track

Agents That Do Real Work

Intended outcome

A working end-to-end demo in which a simulated schema incident is diagnosed through DataHub context, its blast radius is proven through lineage, a safe remediation is proposed, the result is independently verified, and the verified incident knowledge is written back for the next operator or agent.

Built With

Share this project:

Updates