hey, i'm shana.

A full-stack AI application that analyzes contracts, detects red flags, extracts key clauses, and answers questions in plain language using RAG and vector embeddings.

• >Built a RAG pipeline with LangChain and Pinecone, chunking contract documents into 500-token segments (50-token overlap) for semantic search.
• ▸Developed a FastAPI backend with 6 REST endpoints that processes uploaded PDFs, generates automatic contract analysis with red flag detection, and answers natural language questions using conversation memory.
• ▸Used Pinecone for vector search and Supabase for conversation history, messages, and contract metadata.

An AI-powered library assistant for De La Salle Lipa that combines rule-based logic with NLP to help users discover books through personalized recommendations and a real-time catalog.

• >Built a hybrid AI pipeline combining a rule-based engine, TF-IDF + Cosine Similarity NLP layer, and RAG with GPT-4o-mini for intelligent library query handling and personalized book recommendations.
• ▸Separated all AI logic into a FastAPI Python backend serving a Next.js 16 frontend, with Supabase handling real-time catalog availability and Google OAuth restricted to @dlsl.edu.ph accounts.
• ▸Designed a QR-based borrowing and return workflow with automated due date notifications, overdue alerts, and role-based access control for Guest, Student/Faculty, and Librarian users.

An AI-powered recitation coach that helps Filipino students build oral confidence through bilingual speech practice with judgment-free, real-time feedback.

• >Built a Node.js/Express backend with 11 API endpoints, including an AI vision pipeline using Llama 4 Scout (via Groq) to scan printed learning modules through the camera and extract text for practice.
• ▸Implemented four escalating recitation modes — Read-Aloud, Paraphrase, Cold Call, and Stand & Deliver — coached and scored on accuracy, clarity, and confidence using Llama 3.3 70B and Whisper Large V3 Turbo.
• ▸Designed a privacy-first, low-bandwidth experience across 10 screens, with on-device text-to-speech and no image or audio storage.

A full-stack pet care service platform connecting pet owners with service providers across Batangas, Philippines.

• >TODO — describe FurSure's AI feature here (what it does, which model/API it uses).
• ▸Implemented role-based access control for three user types — Pet Owner, Service Provider, and Admin — with Supabase Auth.
• ▸Shipped 25 pages and 15 API routes across the owner, provider, and admin dashboards, backed by 48 reusable components.
• ▸Replaced mock dashboard data with live Supabase queries and integrated document verification workflows for service providers.