← Back to all projects

Hackathon · AI Systems

ORBIT

Jac/JaseciPythonReactGraph AlgorithmsLLM

You feed ORBIT text and it builds a knowledge graph where nodes are concepts, people, and sources, and edges are typed relationships like 'builds upon', 'contradicts', 'inspired by'. Then autonomous AI walker agents traverse that graph to find hidden connections, answer questions via pathfinding, fight knowledge decay using the Ebbinghaus forgetting curve, and map intellectual territory. When you ask ORBIT a question, a Pathfinder agent literally walks from concept to concept through your knowledge, and the path it takes is the reasoning chain. The entire backend is written in Jac, an AI-native programming language where walkers, typed nodes, and typed edges are first-class primitives. 12 LLM-powered functions across 4 modules with semantic prompt annotations and per-function temperature tuning. The Python engine layer handles BFS/DFS pathfinding, PageRank, Louvain community detection, Brandes betweenness centrality, and Ebbinghaus forgetting curve with spaced-repetition reinforcement. 265 passing tests covering graph algorithms, decay model, scoring, Jac compilation, and live walker API integration.

What it does

5 autonomous walker agents: Ingestor, Pathfinder, Explorer, Consolidator, Cartographer

Ingestor extracts concepts via LLM and walks existing graph node-by-node to decide edge creation

Pathfinder answers questions by BFS shortest-path between anchor concepts, synthesizing from the walked path

Explorer runs autonomously every 75 seconds, performing weighted random walks biased toward weak edges

Consolidator detects edges decaying below Ebbinghaus forgetting curve thresholds and flags orphan nodes

Cartographer maps knowledge territory by domain, identifies dense zones, sparse frontiers, and bridge concepts via betweenness centrality

12 by llm() AI functions with semantic annotations and tuned temperature (0.0 factual through 0.7 creative)

Python engine: BFS/DFS, PageRank, Louvain community detection, Brandes betweenness centrality

React frontend with force-directed graph visualization showing real-time walker traversal animations

InsForge (YC-backed) integration: Postgres persistence with per-user isolation and realtime WebSocket sync

265 passing tests across graph algorithms, decay model, Jac compilation, and live API integration

Tech stack

BackendJac (Jaseci) with Python superset interop
AI Functions12 by llm() across extraction, assessment, synthesis, narration
Graph EngineBFS, DFS, PageRank, Louvain, Brandes centrality, Ebbinghaus decay
FrontendReact, react-force-graph-2d, force-directed layout
DatabaseInsForge (Postgres + realtime WebSocket pub/sub)
AuthJWT with per-user graph isolation
HostingVercel (frontend), Render (Jac + API), InsForge (cloud DB)
Testing265 tests (graph algorithms, decay, scoring, Jac compilation, API)
← Back to all projects