I build production AI systems, not demos, not wrappers.
Recent example: a RAG system for an Arabic news portal spanning 10 websites and 1M+ documents. I built scrapers that sync incrementally and recover on their own when a site's layout changes, handled Arabic-specific problems like right-to-left text and dialect variation across sources, and layered hybrid search, reranking, and recency weighting on top of vector retrieval so answers stayed accurate as stories updated. It shipped as a voice-based agent with persona and multi-turn memory, wired end to end into the front end, not just a backend service.
Other things I've built:
- Multi-agent orchestration — specialized agents collaborating on structured research tasks
- GAN-based cyberattack detection — [confirm real accuracy/false-positive numbers before publishing]
- GPT-2 built from scratch — not fine-tuned, not wrapped, built
- Custom embedding architectures — 128-dim vectors, multi-head attention, trained from the ground up
- AI/ML → Python, TensorFlow, PyTorch, Scikit-learn
- LLMs & Agents → Gemini, GPT-4, Vertex AI, LangChain, n8n
- Vector DBs → pgvector, Pinecone, FAISS
- Cloud → Google Cloud Platform (Vertex AI, BigQuery, Cloud Storage)
- Data → Apache Spark, Apache NiFi, ETL Pipelines
- Infra → Docker, REST APIs, Flask, Streamlit
- Deep Learning → Autoencoders (VAE/SAE/DAE), GANs, Transformers
- AI Engineer at Ztudium Ltd, London — building and shipping production AI since Jan 2025
- Ex Innovation/Technical Head at Avinya Club @MIT WPU — mentoring teams, directing projects, running hackathon prep
- Branch Topper — Diploma CS-AI (CGPA 9.25)
- Academic Excellence Award — MIT-WPU Spandan 2026
- Ranked top 1% globally in an international cybersecurity competition (10,000+ participants)
- Hackathon winner — 1st place at Hack-MITWPU 2025 (100+ teams) [add others here if real]
- Building an SEO intelligence dashboard integrating Google Search Console, Analytics, and Ahrefs
- Going deeper on transformer internals and production MLOps
- Writing about AI engineering on LinkedIn
