🚀 AI Engineer | Data Scientist
🎓 MSc Data Science @ EURECOM (France)

I build production-grade AI systems using LLMs, RAG, multi-agent architectures, and scalable MLOps pipelines.

• 🔍 Specialized in LLMs, RAG pipelines, and recommender systems
• 🤖 Building agentic AI systems with real-world impact
• ⚙️ Strong focus on end-to-end ML systems (data → model → deployment)
• ☁️ Experienced with cloud + scalable infrastructure (GCP, AWS, Kubernetes)

Agentic RAG system for recruiter-focused candidate search

• 🧠 GPT-based intent understanding
• 🔎 Hybrid Retrieval (Dense + BM25 + RRF)
• 📊 Cross-encoder reranking + evidence extraction
• ⚡ Real-time system with FastAPI + Node.js + MongoDB

📈 Improved ranking quality (Precision@K, nDCG) over baseline retrieval systems

Generative AI for biological simulation

• 🧬 Conditional diffusion models (DDPM + UNet)
• ⚙️ Optimization via Evolution Strategies vs PPO
• 📊 Achieved ~25% improvement in generation quality (FID)

Unsupervised industrial fault detection

• 🎧 AST + PANN encoders + UNet decoder
• 📊 AUC: 0.943 (+22 pts over baseline)
• 🔍 Detects anomalies without labeled data

Python JavaScript C++ SQL

PyTorch TensorFlow Scikit-learn OpenCV
Transformers Diffusion Models LLMs RAG

LangGraph LangChain AutoGen CrewAI
Multi-Agent Systems Reasoning Models

Docker Kubernetes CI/CD TensorRT
MLflow Airflow DVC

GCP AWS Azure

MongoDB FAISS Pinecone Qdrant

• 🚀 Built large-scale recommender system with ~90% accuracy in filtering irrelevant results
• ⚡ Achieved 8× throughput using GPU optimization (A100, TensorRT)
• 📉 Reduced latency by ~40% in production ML systems
• 📊 Improved prediction accuracy by ~25–30% across multiple domains

• 🥉 3rd Place — Global TiE Hackathon (1400+ teams)
• 💻 275+ LeetCode problems solved

• 🌐 Portfolio: https://adnankarim.dev
• 💼 LinkedIn: https://www.linkedin.com/in/adnan-karim-dl
• 📧 Email: adnankaremjs@gmail.com

I enjoy building systems where AI doesn’t just predict — it reasons, retrieves, and acts.