Full-stack AI application demo

AI answers with sources, not guesses.

A production-style RAG application that searches culinary knowledge, retrieves the most relevant passages, and generates clear answers with visible source evidence. Built to show practical AI engineering, not just a prompt box.

Document retrieval Vector search Grounded answers Cloud-ready backend
For business owners Shows how AI can answer from company documents, manuals, training material, or knowledge bases.
For technical reviewers Demonstrates API design, database modeling, vector search, prompt control, and deployment thinking.
For users Provides a simple interface where answers are readable, sources are visible, and the workflow feels polished.
Technical skill breakdown

Built like a real product.

This page explains the engineering behind the demo from an audience perspective: what problem it solves, what technologies are used, and why the implementation is more professional than a simple AI wrapper.

Core strength

RAG application architecture

The system follows a professional retrieval-first approach: documents are processed into chunks, converted into embeddings, searched by similarity, and then passed into the AI model as grounded context.

✓Uses source material instead of relying only on model memory.
✓Returns page ranges, relevance, distance, and chunk IDs for transparency.
✓Separates retrieval, prompt control, generation, and response formatting.
System flow
User question input Embedding search pgvector Relevant chunks ranked Prompt assembly controlled Answer + sources output
Database

PostgreSQL data modeling

Stores documents, chunks, page metadata, embeddings, and retrieval information in a way that supports scalable search and clear source display.

AI search

pgvector similarity search

Uses vector similarity to find meaning-based matches, allowing the app to retrieve relevant material even when the question wording is different.

Prompt control

Answer mode design

Supports strict, balanced, creative, and recipe modes so the same backend can respond differently depending on the user’s goal.

Deployment

Cloud-ready mindset

Designed with environment variables, service deployment, secret handling, and Google Cloud Run compatibility in mind.

Interactive demo

Ask the kitchen library.

Choose a mode, ask a question, and inspect the retrieved sources.

Ingest Chunk Embed Retrieve Adjust Temperature Answer + Sources

Question

Search cooking theory, techniques, and source pages.

Preparing answer...
Searching culinary sources...
Estimated time: 8–15 seconds
Ready for service. Ask a question to generate a grounded answer with retrieved sources.