Skip to content
Arjit Kulkarni
01Full-stack & AI Engineering

CUTLINE

An AI-assisted permission repair console, with proof built in
Role
Software development — full-stack & AI project
Context
Hackathon project · October 2026
Stack
Next.jsReactTypeScriptReact FlowTanStack QueryFastAPIFalkorDBMCP
Project links
Visit live site ↗cutline-accessgraph.vercel.apparjitkulkarni/cutline-access_graph
01  The problem

A Next.js graph console and permission-repair agent: trace risky access, explore safe repairs, and verify the result before applying a change.

Permissions can reach a user through several inherited roles. Removing one route may leave another open, while removing a shared role can break legitimate work. CUTLINE makes those dependencies visible and checks repairs before they reach the active policy.
02  Approach
  1. 01

    A Next.js console uses React Flow to surface access paths, with dedicated views for repair planning, evidence and regression tests.

  2. 02

    FastAPI and FalkorDB trace inherited access and evaluate repair candidates on isolated graph copies.

  3. 03

    Typed agent tools, also available through MCP, support investigation without giving the model arbitrary database-write access.

  4. 04

    Real HTTP checks compare a proposed repair with independent enforcement. Human approval and revision checks guard application of the change.

03  Outcome

  • An end-to-end workflow connecting visual investigation, repair planning, runtime evidence and approval.

  • Stale evidence prevents application when the underlying policy changes.

  • Exportable regression tests preserve expected access decisions.

Typed agent tools
0Typed agent tools
Shared tool interface
MCPShared tool interface
Runtime verification
HTTPRuntime verification

Concepts applied

  • Interactive graph UI
  • Agent tool design
  • Graph traversal
  • Human approval
  • Verification
  • Full-stack integration
All work