- AI agent
- n8n
- Next.js
- Supabase
CliniDesk
Doctor appointments on WhatsApp. Patients book in chat, and doctors manage appointments and patient history from a dashboard.
Next.js, Supabase, n8nView details about CliniDesk
- AI agent
Nexa AI
Shop by asking, not by filtering.
A complete demo store with natural-language search, a shopping assistant that answers from the catalog, comparison, recommendations and image search. A personal project, built end to end by Agentix founder Ahtasham Naeem.
Role: Full-stack developer
Nexa AI is a complete AI shopping assistant demo store with natural-language search, catalog-grounded answers, side-by-side comparison, recommendations and image search. It is a personal project of Agentix founder Ahtasham Naeem, who built both the Next.js front end and the FastAPI, PostgreSQL and pgvector back end, with MiniLM and CLIP embeddings and optional Google Gemini.
Last updated
- Type
- Personal project (demo store)
- Stack
- Next.js, FastAPI, PostgreSQL
- AI
- MiniLM, CLIP, optional Gemini
- Catalog
- 420 demo products
What it does
The parts of Nexa AI that users use every day.
Natural-language search
Describe what you want in your own words and get matching products.
Grounded assistant
Answers come from the catalog and are checked before they are shown.
Side-by-side comparison
Compare products and read a short summary of the differences.
Image search
Upload a photo and find visually similar products.
How it works
- 1
Search or ask
Type a request or ask the shopping assistant.
- 2
Compare
Put products side by side and read the summary.
- 3
Check out
Place an order in a simulated checkout.
Screens
Open a larger viewSearch by description or by photo. 
Open a larger viewThe assistant answers from catalog records only. 
Open a larger viewProduct cards with stock levels and compare.
Built with
- Next.js
- React
- TypeScript
- Tailwind CSS
- FastAPI
- PostgreSQL
- pgvector
- MiniLM
- CLIP
- Google Gemini
More products
- Web app
- Next.js
- Supabase
- E-commerce
- Local business
Naeem Stationery & General Store
The online shop of a stationery and general store in Sahiwal, with WhatsApp orders and same-day delivery.
Next.js, Supabase, VercelView details about Naeem Stationery & General Store
How the Nexa AI shopping assistant stays accurate
Every generated answer passes a grounding validator before it is shown. The validator checks the cited product IDs, prices, attributes and numbers against the database. A rejected answer is retried once and then replaced by a fallback.
Without a Google Gemini key, the assistant and the comparison summary use non-generative fallbacks: the same retrieval and the same grounded product cards, with a visible notice.
What Nexa AI is built with
The web app uses Next.js 16 with React 19, TypeScript and Tailwind CSS. The API is Python 3.12 with FastAPI and SQLAlchemy. Products live in PostgreSQL 17 with pgvector, with separate text and image vector tables and exact cosine search.
Text is embedded with all-MiniLM-L6-v2 and images with CLIP ViT-B/32. Google Gemini is optional and is used for assistant answers and comparison summaries.
Who built Nexa AI
Nexa AI is a personal project of Agentix founder Ahtasham Naeem, and it is complete. Ahtasham built all of it as a full-stack developer: the Next.js front end, the FastAPI back end, the PostgreSQL and pgvector database, and the search and assistant features on top of them.
It is not a live store. The catalog is demo data and checkout is simulated, so the project shows how an AI shopping assistant works from end to end without taking real orders.
Nexa AI is part of our AI chatbot and AI agent development and custom web development work. Want something similar? Get a quote for your project.
Nexa AI FAQ
Is Nexa AI a live store?
No. It is a demonstration. The catalog comes from the Amazon Berkeley Objects dataset, prices and stock are synthetic values, and payment is always simulated.
Who built Nexa AI?
Agentix founder Ahtasham Naeem, as a personal project. Ahtasham built the whole app as a full-stack developer, from the Next.js front end to the FastAPI and PostgreSQL back end.
How does Nexa AI image search work?
Product images are embedded with CLIP and stored in PostgreSQL with pgvector. A photo is matched against them by cosine similarity.
What happens when the AI model is not available?
The assistant falls back to the same retrieval and grounded product cards without generated text, and it shows a notice.
Who built this
Ahtasham Naeem is the founder of Agentix and a full-stack developer with 5+ years of experience and 20+ projects delivered, building AI agents, automation and web apps with n8n, Make.com, Next.js, React and Ruby on Rails. More about Agentix · LinkedIn · GitHub
Need something like Nexa AI?
Tell us what you need. We will suggest the right approach and a price.