Mohamed Amara

During my 7-month internship, I contributed to the full lifecycle of building, evaluating, and deploying a production-grade RAG solution. • Prototyping & UI: Built an initial prototype with Haystack + Streamlit, later contributing to a modern UI migration using Next.js. • Data Engineering: Processed and cleaned diverse sources (HTML/PDF documentation, Jira issues, internal community). Applied LLM-based summarization, filtering, and preprocessing to generate high-quality knowledge data. • Vector Database & Feedback Loop: Evaluated vector DBs and implemented pgvector to combine relational schema (users, conversations, feedback) with semantic search. Designed a feedback loop to continuously improve retrieval. • Model Experimentation & Fine-tuning: - Ran open-source models locally (RTX 3060, 12GB VRAM) via Ollama, experimenting with Qwen2.5/3 (14B), Mistral, LLaMA, etc. - Scaled to AWS GPU instances (96GB VRAM) to run and fine-tune larger models like Qwen3 32B using Unsloth (LoRA). - Migrated serving from Ollama to vLLM for production performance, leveraging LoRA support. - Deployed a finetuned BGE Large 1.5 EN embedding model via Hugging Face Inference for efficient vectorization. • Evaluation & Monitoring: Implemented evaluation with Ragas, retrieval metrics, and custom RAG pipeline metrics. Built evaluation datasets and dashboards with Prometheus, Grafana, Jaeger, and OTEL to track latencies (embedding, retrieval, LLM generation) using P95/P99 metrics. • MLOps & Deployment: Worked across local, staging, and production environments, improving scalability and observability of the RAG system. Tech stack: Haystack, Streamlit, Next.js, pgvector, Ollama, vLLM, Ragas, Unsloth, Prometheus, Grafana, Jaeger, OTEL, Hugging Face Inference, AWS GPU instances, Qwen, LLaMA, Mistral.

• Developed RESTful APIs using NestJS, TypeScript, and Prisma for a vehicle tracking app, catering to both a Flutter mobile application and an admin dashboard. Enabled features such as real-time GPS tracking, geofencing, speed alerts, historical route playback and reports generation feature on vehicle’s trips and stops. • Implemented Firebase Cloud Messaging to provide users with real-time notifications about speed limits, geofences exit and oil change reminders. • Worked on setting up the CI pipeline, automating and deployment processes, ensuring reliable software releases. • Implemented reports generation feature on vehicle’s trips and stops. • Contributed to Flutter app development.

•Developed and maintained the RESTful APIs using Spring Boot framework, and utilized Git and GitLab for version control and continuous integration/continuous deployment (CI/CD) pipeline. • Designed and implemented MongoDB database schemas, leveraging the aggregate framework to create optimized queries for efficient data retrieval. • Worked with the team to leverage SonarQube for code quality analysis, reviewing the reports and implementing code fixes and improvements to meet quality standards. • Integrated SonarQube for code quality analysis, and implemented code fixes and improvements to meet quality standards. • Utilized agile methodologies and tools such as JIRA and Trello to manage project tasks and timelines

• Conceived, designed and implemented an e-learning website for a private school. • Developed a learning management system and integrated an online conferencing solution (BigBlueButton) • Used the PERN stack (PostgreSQL, ExpressJS, ReactJS, NodeJS), BigBlueButton

-Implemented , designed and conceived a Real-time in-App chat -Stack used : MERN Stack -Used Redux for global state management -Used Socket.Io for exchanging the messages in real-time -Git & Github for version control