WorkIn1 - AI Digital Identity & Journey Intelligence

WorkIn1 turns a student's scattered documents — certificates, resumes, project reports, internship letters, portfolio links — into an intelligent digital identity that understands their journey rather than just storing it. Originals are never modified; every result links back to the untouched source file. Pipeline: files are parsed (PyMuPDF for PDFs, Tesseract OCR for scanned certificates, python-docx, readability extraction for URLs), auto-categorized by an LLM into Projects / Skills / Certifications / Internships / Achievements / Academics, then chunked and embedded locally into a persistent ChromaDB store. Four capabilities sit on that foundation: 1. Hybrid Semantic Retrieval (RAG) — dense vector search fused with BM25 keyword search via Reciprocal Rank Fusion, so "show my AI projects" surfaces a project titled "Neural Network Classifier" while exact organization names still match reliably. Each result carries a grounded match explanation derived from the actual retrieved chunk and shared entities, plus a cited RAG answer. 2. Knowledge Graph — an NLP relationship engine extracts entities and relationships per document, then canonicalizes them with fuzzy deduplication so "Python" and "python programming" merge into one node. The result is a traversable graph: Certification → Skill → Project → Internship → Career Path, with cross-document paths spanning the full journey. 3. Journey Timeline — auto-dated, year-grouped growth view linking back to originals. 4. Journey Insight — the system reasons over the whole graph to infer the student's likely career trajectory, identify strengths, and name concrete gaps. This is what separates it from storage: it doesn't just hold the past, it reads the direction. Techniques: embeddings, semantic search, hybrid retrieval, RAG, vector database, NLP entity/relationship extraction, knowledge mapping, fuzzy entity resolution. Fully free and local — local embeddings (all-MiniLM-L6-v2), Groq with Ollama fallback, no paid API required. One command to run; ships with an 11-document demo journey so it's instantly explorable. Retrieval accuracy measured on a 15-query evaluation set (see eval/ and README). The defining moment: "I never have to search through folders again — and it understands where I'm headed."