Case 05
AI / Machine Learning
Retrieval-Augmented Generation (RAG) System
Grounding large language models in retrieved, traceable context.
The project
Engineered a custom RAG pipeline integrating large language models with an external vector database to support accurate, traceable and up-to-date data synthesis.
Optimised context retrieval mechanisms to mitigate model hallucinations and improve system response latency.
Specification
- Category
- AI / Machine Learning
- Context
- Large Language Models
- Retrieval-Augmented Generation
- External Vector Database
- Context Retrieval
- Index
- 05 / 06
The pipeline
- 01
User query
A question arrives in natural language and is prepared for search.
- 02
Retrieval
The query is matched against an external vector database to find the most relevant passages.
- 03
Context
Retrieved passages are ranked and assembled into focused, up-to-date context.
- 04
LLM
A large language model generates an answer grounded in that context rather than in memory alone.
- 05
Response
The answer comes back traceable to its sources — the step that keeps hallucination in check.