All Projects

Qanouny AI-Powered Legal Assistant

(DEPI Pioneer Graduation Project)

AI Team Leader

  • Led the AI team and collaborated cross-functionally with Data Engineering and Flutter teams to design and deliver a production-ready RAG-based legal AI system.
  • Architected an end-to-end multi-modal AI pipeline (text, audio, image, PDF), integrating OCR, speech-to-text, embeddings, and LLM reasoning into a unified backend.
  • Designed and implemented a vector-based legal retrieval layer in coordination with the Data Engineering team, ensuring scalable, source-grounded, and auditable legal responses.
  • Worked closely with the Flutter team to align AI outputs with user experience requirements, including risk indicators (Red/Amber/Green), simplified legal language, and explainable responses.
RAG OCR Speech-to-text Embeddings LLM Reasoning Groq API FastAPI
Live Demo

CorpGuideAI - Intelligent HR Policy Assistant

  • Developed a production-ready RAG application to automate HR inquiries, utilizing LangChain and Llama 4 (via Groq) to achieve sub-2-second inference latency.
  • Implemented Semantic Chunking to optimize vector retrieval accuracy for bilingual policy documents, significantly outperforming traditional fixed-size chunking methods.
  • Engineered a robust microservices architecture by decoupling the FastAPI backend (Dockerized on Hugging Face) from the frontend (Vercel), featuring a self-healing database mechanism for error resilience.
LangChain FastAPI RAG Groq API Semantic Chunking Hugging Face Vercal
View Code Live Demo

Customer Churn Prediction

  • Built and evaluated a customer-churn prediction pipeline using models like LightGBM, XGBoost, and RandomForest; LightGBM achieved AUC ≈ 0.836 and recall ≈ 0.82, proving strong ability to detect likely churners.
  • Implemented customer segmentation via K-Means clustering (integrating churn probability as a feature) to identify high-risk segments and support targeted retention strategies.
  • Developed an interactive dashboard (with Streamlit) that visualizes churn risk and cluster- level metrics, enabling clear actionable insights for business decision-making (e.g. retention campaigns).
Machine Learning Streamlit Plotly
View Code View Notebook

Smart Inventory & Sales Forecasting System

  • Engineered an automated ETL pipeline to transform raw retail data into a normalized SQLite database, implementing rigorous data cleaning rules to ensure high-quality inputs for modeling.
  • Developed a monthly demand forecasting engine using Random Forest; utilized MLflow for experiment tracking and dynamic champion-model selection, achieving an $R^2 \approx 0.80$.
  • Architected a secure Text-to-SQL AI Agent using LangChain and Llama 4 scout (Groq), enabling stakeholders to query database insights via natural language with strict read-only security guardrails.
  • Deployed an interactive Streamlit dashboard integrating real-time forecasting and the AI assistant, delivering actionable business intelligence through dynamic Plotly visualizations.
LangChain RAG Groq API Streamlit MLflow SQLite SQL-Agent Plotly
View Code Live Demo
Back to Home