An AI-powered Arabic literature platform, built as a graduation project by a team of four.
- Role
- Application Services Team Lead
- Year
- 2024 — 2025
- Focus
- AI · Full-stack · Team leadership
LiteraryHub
The problem
An Arabic literature platform where authors upload their work and readers can read it or listen to it, with AI handling the parts that are slow or hard to do by hand.
Role and scope
- Graduation project, grade A+
- Co-led a team of 4 as Application Services Team Lead
- Owned audio streaming, the author upload pipeline, integration tests and QA sign-off across 5 GitHub repos
- Audiobook streaming on MongoDB GridFS and Node.js
- Integrated 5 AI models: SDXL Turbo (covers), JAIS 13B (Arabic restricted-topic detection), T5-Small (summarization, Rouge F1 0.95), MongoDB vector search (plagiarism), gTTS (audiobooks)
- Survey (n=12): 92% positive sentiment, 94% recommend
Architecture
Key decisions
- 01
Treat models as a pipeline
Five models each own one job: SDXL Turbo for covers, JAIS 13B for Arabic restricted-topic detection, T5-Small for summaries, MongoDB vector search for plagiarism, and gTTS for audiobooks.
- 02
Gate publishing with a model
The author upload pipeline runs an AI restricted-topics detection model end to end before a work goes anywhere.
- 03
Stream audio natively
Audiobooks stream from MongoDB GridFS through Node.js, so listening starts without downloading a whole file.
- 04
Own quality, not only code
Owning integration tests and QA sign-off across 5 GitHub repos kept a four-person team shipping the same product.
Outcomes
- Graded A+ as a graduation project
- Summarization scored Rouge F1 0.95
- Survey (n=12): 92% positive sentiment, 94% recommend
Stack
- Node.js
- MongoDB
- GridFS
- SDXL Turbo
- JAIS 13B
- T5-Small
- gTTS