Applied AI / Search Platform
AI Company Search & Expert Discovery Platform
Built a live UK-market AI search platform for company and expert discovery, combining RAG, document parsing, vector retrieval, multi-LLM selection and payments.
Why this work matters
Business context
Built a live UK-market AI search platform for company and expert discovery, combining RAG, document parsing, vector retrieval, multi-LLM selection and payments.
Technical risk
The product needed to help users find companies and experts through meaning, documents and context rather than brittle keyword matching. It also had to support production payment flows and multiple LLM providers.
Engineering outcome
The product became more than a chatbot: it behaved like a professional search engine where retrieval quality, document understanding, payments and user trust all mattered.
The Product Challenge
The product needed to help users find companies and experts through meaning, documents and context rather than brittle keyword matching. It also had to support production payment flows and multiple LLM providers.
The hard part was designing an AI system that could parse documents, retrieve the right context, choose the right model path and still behave predictably enough for a live business product.
The product became more than a chatbot: it behaved like a professional search engine where retrieval quality, document understanding, payments and user trust all mattered.
My Engineering Contribution
- Designed the RAG pipeline using MongoDB for source content and Qdrant for vector retrieval, with document parsing before embedding and retrieval.
- Built a Spring AI and FastAPI service combination so JVM product services and Python AI workflows could evolve cleanly.
- Implemented multiple LLM selections with provider-specific keys, allowing the system to route work across model options instead of depending on one model.
- Integrated Stripe escrow and payment workflows around the search/discovery product experience.
- Worked on production concerns including parsing quality, retrieval ranking, latency, cost control and deployment readiness.
System & Product Considerations
- Document parsing and chunk quality before embedding
- MongoDB source data plus Qdrant vector indexes
- Spring AI orchestration with FastAPI AI services
- Multiple LLM providers and API-key management
- Stripe escrow, payment states and product trust
Technical Areas
What This Project Taught Me
- A production AI product is a search, data and payments product before it is an LLM demo.
- RAG quality is won in parsing, chunking, retrieval and ranking. The final model call is only one step.
- Multi-model support is useful only when keys, cost, failure modes and routing are designed as product concerns.
Need someone to own this kind of technical complexity?
Let's talk it through.
muhammadmansoor417@gmail.com