I build ML systems from scratch, ship them end-to-end, and measure whether they actually work.
🌐 Full portfolio & case studies → samielakkad.github.io
- 🎓 AI MSc @ Tsinghua SIGS · 清华大学 — GPA 3.9 / 4.0 · full CSC scholarship
- 🏅 President's Scholarship @ NWPU · 西北工业大学 — full award (CS / AI), final grade 87%
- 🥉 Baidu · 百度 ERNIE Hackathon 2025 — 3rd Prize (top 3 of 23 finalists, 170+ submissions)
- 🏆 Winner — Tsinghua AIID Yearly Competition (NetDragon · 网龙 partner track)
- 🚗 Built a YOLOv8 detector from scratch for autonomous-driving perception @ Neusoft · 东软集团
- 🚀 Shipped a live LLM product end-to-end, solo — jak.ma: multi-provider routing, tool-calling agents, Darija NLP
- 🌍 5 languages — Arabic · Tamazight · French · English (IELTS 7.5) · 中文 (HSK 4, 90/100)
I'm an engineer who ships. I trained as a computer scientist at NWPU · 西北工业大学 (President's Scholarship), I'm doing my MSc in Artificial Intelligence at Tsinghua · 清华大学 SIGS, and along the way I've built object detectors from scratch, evaluated frontier LLMs at Baidu · 百度, and single-handedly shipped a live, LLM-powered product serving real users.
My edge is the full loop in one person: build the model → ship the product → prove it works. That range — across applied-AI engineering, forward-deployed delivery, and product — plus five languages and a bias for shipping, is what makes me hard to box into a single lane (and easy to plug into many).
| 🧱 Build | Implement modern ML from first principles — YOLOv8 detection, evolutionary optimization algorithms, LLM routing & agents. Not library wrappers. |
| 🚀 Ship | Take systems end-to-end to production — solo-built jak.ma, a live LLM-powered service marketplace. Data → model → deploy → debug in the field. |
| 📏 Evaluate | Measure model quality rigorously — a multi-dimensional evaluation rubric carried from Baidu ERNIE to my own products. I care whether a model is good, not just whether it runs. |
Work built for companies and institutions — each links to the organisation in English · 中文.
| Project | Built for | What it is |
|---|---|---|
| YOLOv8 from scratch | Neusoft · 东软集团 | Autonomous-driving road-object detector, hand-built in PyTorch — CV internship, 2024 |
| ERNIE evaluation notes | Baidu · 百度 | Model-evaluation methodology from the ERNIE (文心) Mentor Program — 2025 |
| Sovereign School OS | NetDragon · 网龙 × Tsinghua · 清华大学 | AI-native K-12 platform — winner, Tsinghua AIID Yearly Competition (NetDragon partner track) |
| ADCMOplus | NWPU · 西北工业大学 | Graduation research — online-learning evolutionary algorithm for dynamic constrained optimization |
🤖 Computer Vision · 🧠 LLMs & Agents · 🗣️ NLP · 🧬 Optimization · 🎓 EdTech · ⚖️ LegalTech · 📦 Product
- yolov8-from-scratch-pytorch — full YOLOv8 (CSPDarknet + PANet + DFL head) built, trained, evaluated & debugged end-to-end. Autonomous-driving perception @ Neusoft 东软集团.
- jak-ma-engine — multi-provider LLM routing + tool-calling agent powering a live product.
- jak-ma-eval-suite — evaluation methodology for a two-pass grounded-retrieval system.
- ernie-evaluation-notes — 5-dimension LLM evaluation rubric. Baidu 百度 ERNIE Mentor Program.
- jakma-darija-classifier — production Darija (Moroccan Arabic) intent + city classifier, Hugging Face–ready.
- darija-nlp-resources — curated corpora, papers & tools for low-resource Darija NLP.
- ADCMOplus — novel online-learning evolutionary algorithm with a proven regret bound; beat three SOTA baselines. Graduation research @ NWPU 西北工业大学.
- Sovereign School OS — AI-native K-12 platform, 50+ React/TS components, FERPA-compliant by construction. AIID winner — NetDragon 网龙.
- hsk-master-ai — AI Mandarin-exam tutor built around the retake (writing diagnostics, prosody scoring).
- tsinghua-aiid-2025-team-docs — interactive docs site; Best Documentation Award, Tsinghua AIID Hackathon 2025.
- mirath-morocco — Moroccan inheritance-law solver (Faraid, Asaba, Awl, Radd) with parcel-level land valuation.
- jak-ma-case-study — production case study: verifier-gated retrieval shipped into a live marketplace.
- pm-frameworks-darija — reusable PM frameworks for shipping LLM products in low-resource markets.
Applied AI · Engineering & Product · Ships end-to-end — building, shipping, and measuring AI systems.