FalkorBrain: Autonomous Engineering Context Engine

Team: Team 7vengers Tracks: Track 01 (Agents That Act on Connected Data), Track 02 (Agent Memory and Coordination), Track 03 (Company Brain) Repository: https://github.com/krishsf5/Graph-Hack-Hackathon-Team-7vengers Live Demo: https://falkorbrain.onrender.com Elevator Pitch FalkorBrain turns AI agents from isolated vector searchers into relationship-aware systems. It uses FalkorDB as a central Knowledge Graph context layer that connects team members, architectural decisions (ADRs), code dependencies, and multi-tier memories (episodic, semantic, procedural). This enables multi-hop reasoning, transparent explainability traces, and MCP tool orchestration. The Problem Standard AI agents lack relational context: They treat code and memories as flat text chunks, missing multi-hop dependencies. They cannot explain the chain of reasoning behind a decision. They lose organizational context: who authored a decision, why a rule exists, and which modules depend on it. How FalkorDB Shapes the Agent Multi-Hop Traversal: The agent answers questions with parameterized Cypher queries that trace paths such as (User)-[:AUTHORED]->(Decision)-[:AFFECTS]->(SourceFile)-[:IMPORTS]->(DependentFile). Explainability Trace: Every answer cites the exact path traversed in FalkorDB, so users can see how it was derived. Model Context Protocol (MCP) Integration: Exposes graph querying, impact tracing, and memory retrieval tools to any MCP-compatible agent or IDE. Dynamic Codebase Ingestion: Scans JavaScript and TypeScript with the Babel AST parser and maps files, imports, and symbols as live nodes and edges in FalkorDB. Persistent Memory Across Sessions: Stores episodic meeting notes, semantic concepts, and procedural review rules as graph nodes. Tech Stack Graph Database and Context Engine: FalkorDB (Cypher, node and range indexing, graph algorithms) Backend and Orchestration: Node.js, Express, @babel/parser, @babel/traverse, falkordb SDK Frontend Dashboard: React, Vite, vanilla CSS design system Protocols and Standards: Model Context Protocol (MCP), REST API Track Requirement Matrix Track 01: Agents That Act on Connected Data Multi-Hop Reasoning: The agent traces (Decision)-[:AFFECTS]->(File)-[:IMPORTS]->(Importer) and (File)-[:DEFINES]->(Symbol) to map multi-hop blast radius. Graph Queries as Agent Tools: Parameterized Cypher tools (executeCypher, getDecisionImpactTrace, explainConnectionPath) run directly against FalkorDB. Tool Selection via MCP: MCP endpoints expose falkordb_cypher_query, falkordb_trace_decision_impact, falkordb_explain_connection, and falkordb_get_agent_memory. Actions Based on Graph Topology: When a user asks about modifying token.util.js, the agent warns that ADR-001 enforces userCode immutability on that module. Explainable Results: The UI shows breadcrumb chip paths such as [Krish] -[:AUTHORED]-> [ADR-001] -[:AFFECTS]-> [token.util.js], showing the exact path traversed. Track 02: Agent Memory and Coordination Episodic, Semantic, and Procedural Memory: Episodic: Retains meeting takeaways (e.g., Session #101 token revocation rules). Semantic: Preserves durable best practices (e.g., the JWT numerical ID principle). Procedural: Enforces code review rules (e.g., test regression guidelines). Shared State and Multi-Agent Handoffs: Models collaboration with (:Agent)-[:HANDED_OFF_TO {task, status, handoffTime}]->(:Agent) and persists coordination state in FalkorDB. Tenant and User Scoping: Memories and sessions carry tenantId and userId attributes for strict tenancy isolation. Track 03: Company Brain Docs, Tickets, and Chats in One Graph: Connects Jira tickets (SEC-402), engineering specs (DOC-88), chat sessions (Session #101), and ADRs in a single graph. Ownership and Expertise: Explicit (:User)-[:OWNS]->(:SourceFile) and (:User)-[:EXPERT_IN]->(:Concept) relationships answer who owns what and who knows what. Tracing Decisions to Source: Cypher queries trace any decision back to its originating ticket (-[:RESOLVES_TICKET]->) and specification (-[:CITES_DOCUMENT]->). Answers with Clear Sources: Responses cite the exact authors, reviewers, ticket IDs, and URLs from the graph.