Rahul Vadher

Mosaic

# 🏆 Reviewer & Judge Notes: Mosaic AI Welcome to **Mosaic AI**! Thank you for taking the time to review our hackathon submission. To help you evaluate our project efficiently, we’ve put together this quick guide highlighting our core innovations, technical achievements, and how to navigate the project. --- ## 🎯 The Vision: What is Mosaic AI? **Mosaic AI is an AI-powered Digital Identity Platform.** In today's world, our professional identity is scattered across resumes, PDFs, code repositories, and online profiles. Mosaic AI solves this by ingesting your unstructured professional data, extracting entities and relationships using Local Large Language Models, and visualizing your entire career journey as an interactive **Knowledge Graph** and **Chronological Timeline**. --- ## 🚀 Key Technical Achievements (Why this stands out) 1. **Advanced Multi-Database Architecture** Instead of forcing all data into one paradigm, we used the right tool for the job: - **PostgreSQL (Prisma):** Core relational data and application state. - **Neo4j (Graph DB):** Powering the relationship mappings between skills, experiences, and projects. - **Qdrant (Vector DB):** Storing high-dimensional embeddings for semantic search (RAG). 2. **Local AI Inference (Privacy-First Extraction)** Rather than relying solely on costly OpenAI APIs, we built a custom extraction engine using **Ollama (LLaMA3)**. Our backend pipeline parses PDFs/text, structures the data locally, and generates embeddings using **Nomic-Embed-Text**. 3. **Premium UI/UX Engineering** We didn't just build a backend; the frontend is a highly polished **Next.js 15 (React 19)** application. It features a custom "Glassmorphism" design system, dark mode, smooth framer-motion animations, and a highly interactive node-based visualization using **React Flow**. 4. **Enterprise-Grade Backend in a Hackathon** Built with **NestJS**, the backend utilizes a Modular Monolith architecture. It includes robust abstractions, DTO validation, custom interceptors, and a highly scalable folder structure ready for background job queues. --- ## 🧪 The "Golden Path" (What to test during the demo) Since this is a hackathon prototype (v0.1.0), certain features are still in development. Here is the best path to experience the power of Mosaic AI: 1. **The Dashboard (`/dashboard`)**: Check out the premium UI, data visualizations, and metric cards summarizing the user's digital identity. 2. **The Knowledge Graph (`/graph`)**: Interact with the React Flow graph. This is where Neo4j shines, mapping out how a user's skills, companies, and projects interconnect. 3. **The Timeline (`/timeline`)**: View the chronological timeline that the AI built by extracting timestamps and events from the uploaded documents. 4. **The Knowledge Explorer (`/knowledge`)**: See the raw extracted entities and documents that power the graph and timeline. --- ## 🚧 Current Limitations (Hackathon Scope) We achieved about **65%** of our total ambitious vision over the hackathon. You might notice: - **Asynchronous Queues:** Currently, the LLM extraction runs synchronously. In production, we plan to move this to a BullMQ/Redis worker queue to prevent HTTP timeouts. - **Pending UIs:** The Upload flow and Chat/Search UIs are fully mapped in the backend but lack the frontend wire-up. - **Auth Bypass:** Supabase Auth is configured but currently bypassed to make local testing easier for reviewers. --- ## 🔮 Future Roadmap If we continue building Mosaic AI, our immediate next steps are: - Enabling the full **AI Chat Interface** (RAG) so users can ask questions like *"Based on my resume, am I qualified for a Senior React role?"* - Implementing **Career Intelligence Insights** to forecast career trajectories and recommend learning paths based on graph gaps. - Full deployment to Vercel (Frontend) and Railway (Backend / DBs). --- Thank you for reviewing **Mosaic AI**! We hope you enjoy exploring the codebase as much as we enjoyed building it.