MemoryVerse AI '26

LifeGraph AI addresses a real problem: students accumulate certificates, project reports, and internship letters scattered across drives and emails, with no way to see how these documents connect or what career story they collectively tell. Rather than building another file manager, the project treats uploaded documents as raw material for an AI system that reads, understands, and connects them — every document, regardless of type, flows through the same pipeline: text extraction (native parsing with an OCR fallback via Tesseract for scanned files), AI-driven structured extraction using a fixed JSON schema, automatic categorization, and embedding for semantic search. This uniformity is what lets the knowledge graph and RAG-based chat search work consistently across resumes, certificates, and reports without type-specific logic. The knowledge graph itself is built from genuine relational signals — shared source documents and technology-name overlap — rather than a hardcoded ontology, and the AI search feature grounds every answer in cited source documents so responses are traceable rather than free-floating. The system is scoped deliberately as an MVP: SQLite and ChromaDB are ephemeral on Render's free tier, meaning data resets on redeploy (acceptable for a live demo, but a production version would need PostgreSQL and a persistent vector store), and the skill-to-entity linking in the graph relies on name matching and shared-document heuristics rather than deep NLP disambiguation. These are known, intentional trade-offs made to prioritize a working end-to-end demo within the hackathon timeline. Technologies & Tools