Orion

Setup Steps (Local) npm install Create .env with GEMINI_API_KEY and Supabase keys (URL, Publishable, Service Role), As Like .git.example in GitHub Repo npm run dev (Live version available via Vercel link). Auth has been intentionally bypassed for frictionless testing. Simply upload a resume, certificate, or paste a GitHub link to see the AI instantly parse and categorize it. 1. NLP & AI Categorization : I integrated Gemini 2.5 Flash Multimodal to instantly "read" uploaded PDFs/Images. It uses Natural Language Processing (NLP) to understand the document's context, extract hardcore technical skills, generate summaries, and classify the document automatically (Module 1 & 2). 2. Embeddings & Vector Databases : After NLP extraction, we generate a 768-dimensional Gemini Vector Embedding for every document. These embeddings are stored in a Supabase pgvector database, enabling true Semantic Search (RAG)—meaning the system understands the meaning of queries, not just keyword matching (Module 5). 3. Knowledge Mapping & UX : Extracted skills are mathematically mapped into an interactive Knowledge Graph, allowing reviewers to visually trace how a specific certification connects to specific skills and projects (Module 3). The AI also automatically extracts dates to generate a Digital Journey Timeline without any manual user input (Module 4). 4.Architecture: Built with React, TypeScript, Tailwind, Gemini Flash & Embeddings, Supabase (pgvector & Storage), and Vercel. We engineered a custom NodeJS buffer pipeline that drops AI image-encoding overhead from 15s down to 0.01s, allowing heavy Multimodal NLP processing to run flawlessly on Vercel's serverless edge.