Discover how iHailey combines multi-hop Knowledge Graph traversals, high-dimensional vector search, 4-tier memory, and 15 autonomous domain AI agents orchestrated via LangGraph.
Deterministic intent decomposition, parallel routing, and task graph execution
Every request entering iHailey is analyzed by the Supervisor Agent. Rather than executing simple single-turn prompts, the Supervisor decomposes complex prompts into structured subtasks (`TaskPlan`) and routes them through a compiled LangGraph State Graph.
Evaluates user intent, complexity, research requirements, and memory needs.
Routes tasks to specialized agents (Coding, Science, Writing, Finance, Legal, etc.).
Audits logic, verifies citations, and eliminates hallucinations before final response.
Multi-hop relationship traversal fused with Pinecone vector search
Traverses Firestore entity-relationship nodes (People, Companies, Papers, Technologies) up to 3 hops deep.
Pinecone vector similarity search with 1536-dimensional OpenAI embeddings for semantic paragraph matching.
Collection, tag, author, date, and project-level permission filters enforced at query time.
Injects persistent user facts, preferences, past project decisions, and active workflows.
Reciprocal Rank Fusion (RRF) merges vector scores, graph weights, and memory facts into a compressed, high-density prompt block.
Universal document parsing, audio overviews, notebook execution & presentation decks
PDF, Word, Excel, PPTX, CSV, JSON, Markdown, YouTube & Web scraper
Two-host AI podcast discussions generated via OpenAI Text-to-Speech
Sandboxed Python & SQL analytical execution with charts & statistics
Mind maps, presentations, study guides, flashcards, and quizzes
iHailey was engineered and developed by Chancellor Minus, Founder and CEO of ChanceTEK LLC and iSynera LLC.