Ameya Bhagat

Keepsake : AI-Powered Career Intelligence

How to Test the Deployed App: 1. Sign Up/Log In: Create a new account or log in to access the dashboard. 2. Upload a Document: Go to the Upload section and upload a PDF/DOCX document (like a resume, internship letter, or certificate). 3. Explore the Extracted Intelligence: - Knowledge Graph: Look at the visual network representing your extracted skills, organizations, and their relationships. - Career Timeline: View the chronologically structured events generated automatically from dates found in your document. - Semantic Search & AI Assistant: Search for terms or ask the assistant questions about your uploaded documents (e.g., "What projects did I work on?"). - Resume Analyzer: Compare another resume file against your uploaded career database to find gaps. Key Design Decisions: - Fast AI Inference: Utilized Groq API for Llama-3 models (70b for text/reasoning and 11b-vision for images) to achieve 300+ tokens/second, enabling real-time metadata extraction from uploaded career documents. - Vector Embeddings & Search: Implemented Google Gemini (text-embedding-004) for generating embeddings and stored them in Supabase using the pgvector extension to power the semantic search and AI Assistant. - Client-Side Text Extraction: Used pdfjs-dist and mammoth to extract raw text directly in the browser before sending it to the AI pipeline, keeping the backend light.

Q-Cure: Zero-Wait Digital Triage & Real-Time Queue Orchestration

Reduced patient registration from 3 minutes to 15 seconds via self-check-in QR kiosks, with 0ms sync latency across clinic dashboards.

3m → 15sPatient Registration Time