JoeAIWorks
Joseph Rajasekar

AI products that show their working.

I'm Joseph Rajasekar, an IT delivery leader with enterprise experience at American Express and a Certified IBM AI Product Manager. I design, build and deliver AI solutions end to end – and each demo here shows, step by step, exactly how the AI reaches its answer.

Projects you can explore

Each project comes with a narrated guided tour, a private live workspace to test with your own data, and read-only access to the code.

Live demo available

AI Assistant & Support Agent for AMS

A shop-management portal with two AI agents: one helps staff do their work, one helps IT fix problems.

AMS runs alteration shops: orders, garments, invoices, payments and customers across branches. The AI Assistant answers staff questions and prepares new orders, status changes and customers for them to confirm. The Support Agent investigates issues in the code, database structure and logs, and proposes reviewed fixes. Both use the person's own sign-in and permissions.

ClaudeAI agentsTool callingPHP + PostgreSQLFastAPI
  • Assistant prepares changes; people confirm
  • Access rules applied in code before the AI sees data
  • Support Agent with read-only tools and cited evidence
  • Claude, with automatic switch to a self-hosted model (Ollama)
Guided tourOn request
Try it with your own dataOn request
Review the codeOn request
A private link, valid for 7 days
Live demo available

AI Talent Search

Find the best candidates in your own network for every new vacancy – with evidence.

Recruiters upload CVs and interview reports. Claude extracts structured profiles, analyses each vacancy into mandatory and preferred requirements, and checks candidates against every requirement with word-for-word evidence. Plain-language search runs locally.

Claude APIRAGHybrid searchPythonPrivacy by design
  • Three-stage matching: search → AI evidence check → transparent score
  • Hybrid search (local embeddings + BM25)
  • Every AI claim backed by a verified quote
  • Personal data pseudonymised before AI
Guided tourOn request
Try it with your own dataOn request
Review the codeOn request
A private link, valid for 7 days
Live demo available

Grounded Document Assistant

Ask questions about your PDFs and get answers with the document and page – or an honest “I don't know”.

Upload PDFs such as policies, manuals or contracts. Each one is read page by page and indexed in PostgreSQL with pgvector. Questions are answered by OpenAI through LangChain from the retrieved passages only, with the document name, page and a verified quote for every fact. When the answer isn't in the documents, it says so instead of guessing.

OpenAILangChainRAGpgvectorFastAPI
  • Hybrid search: pgvector + PostgreSQL full-text
  • Citations with document, page and highlighted quote
  • Says “I don't know” before the AI is called
  • FastAPI /chat, Docker and an evaluation report
Guided tourOn request
Try it with your own dataOn request
Review the codeOn request
A private link, valid for 7 days
Demo coming soon

AI Invoice Extractor

Turn invoices in any layout into clean, validated data – ready for your finance system.

Upload PDF or scanned invoices. Claude reads supplier, dates, totals, VAT and line items, each linked to where it appears on the page; totals are checked automatically and anything uncertain goes to a quick review screen before export.

Claude visionDocument AIValidation rulesPython
  • Any invoice layout, no templates
  • Automatic total and VAT checks
  • Human review of low-confidence fields
  • Export to Excel / CSV / API
Guided tourComing soon
Try it with your own dataComing soon
Review the codeComing soon
A private link, valid for 7 days

In development

Next in the showcase.

Incident Resolution Automation

Tickets from Slack and email classified, routed and resolved with AI.

Elderly Care & Monitoring

A mobile app that helps families and carers look after elderly relatives.

How a demo works

Ask for access and you receive a private link, valid for 7 days.

Watch the guided tour

A short narrated walkthrough of the real application. A side panel explains each step: what data is used, which AI model is called and why.

Try it with your own data

Your own private workspace with sample data loaded. Upload your documents and see the results, with the same explanation panel. Everything is deleted when access ends.

Review the code

Read-only access to the source code on GitHub, so your technical team can check the architecture and quality.

Request demo access

Tell me what you'd like to see and I'll send you a private link, usually within one working day.