Leeds Beckett University and the Saudi AI competition · LLM application · 2024
AIassistantforUmrahfor Ministry of Hajj and Umrah, Saudi Arabia
1st place of about 1,000 entrants in Saudi Arabia's AI competition, April 2024.
1stOf about 1,000 entrants
01 / 03
A question arrives
A pilgrim asks about a step of Umrah.
02 / 03
Retrieve what's relevant
Corrective RAG pulls the closest sources, checks them, and searches the web when they fall short.
03 / 03
Answer with confidence
The fine-tuned model answers from what it found. It took 1st place out of about 1,000 entrants.
The project
- Client
- Ministry of Hajj and Umrah, Saudi Arabia
- Role
- MSc Data Science, Leeds Beckett University and the Saudi AI competition
- When
- 2024
- Discipline
- LLM application
- Tools
- Fine-tuning
- RAG
- Prompt engineering
- OpenAI API
- Streamlit
The situation
Pilgrims need instant, reliable answers about Umrah procedures.
What I built
- A fine-tuned GPT-3.5 assistant with a ChatGPT-like interface, deployed on Streamlit
- For the MSc dissertation: fine-tuning, prompt engineering and corrective RAG with real-time web scraping
What changed
1st place among 15 finalists from 15 countries, out of about 1,000 participants worldwide. The government is considering implementation, with potential investor funding. The dissertation was marked distinction without revision.
- 1stOf about 1,000 entrants
Hiring for marketing measurement or customer ML in the UK?
Eligible to work in the UK, no sponsorship needed. Based in London.
The CV is a two-page PDF, rebuilt from this site whenever it changes.
Jihar Gifari · London · jigifari@gmail.com · +44 7985 687426