class AIEngineer:
def __init__(self):
self.name = "Sheheryar Ramzan"
self.role = "Full Stack AI Engineer"
self.education = "BS Computer Science @ FAST-NUCES (GPA: 3.49/4.0)"
self.location = "Pakistan 🇵🇰"
self.email = "sheheryarramzan01@gmail.com"
self.experience = {
"AlphatechLogics" : "AI Engineer + Automation Engineer (Mar 2025 – Present)",
"Software Motion" : "Perception AI Engineer (Jul 2024 – Present)",
"CoE-AI" : "Machine Learning Intern (Jul 2023 – Aug 2023)",
"FAST-NUCES" : "Teaching Assistant – Numerical Computing (Jan–Jun 2024)",
}
self.expertise = [
"Computer Vision", "Deep Learning", "LLMs", "RAG",
"ONNX Deployment", "Generative AI", "3D Reconstruction"
]
self.awards = [
"🏆 Dean's List — 5 Semesters",
"🥉 Bronze Medal — 3rd Position in Batch (Semester 7)"
]
def say_hi(self):
print("Let's build intelligent systems that push the boundaries 🚀")
me = AIEngineer()
me.say_hi()| 🏢 Company | 💡 Role | 📅 Duration | 📍 Location |
|---|---|---|---|
| AlphatechLogics | AI Engineer + Automation Engineer | Mar 2025 – Present | Lahore · Remote |
| Software Motion (Suzhou) | Perception AI Engineer | Jul 2024 – Present | China · Remote |
| CoE – Artificial Intelligence | Machine Learning Intern | Jul 2023 – Aug 2023 | Islamabad, PK |
| FAST-NUCES | Teaching Assistant – Numerical Computing | Jan 2024 – Jun 2024 | Islamabad, PK |
🔹 AlphatechLogics
├─ Built real-time DensePose + YOLO logo transfer pipeline → 50% jitter reduction
├─ Voice-to-SQL pipeline (Mandarin + English) with 95% query accuracy
└─ Retail analytics POC: age, gender & dwell-time estimation → 93% detection precision
🔹 Software Motion (Suzhou)
├─ Team Lead for ONNX deployment — model export, PyTorch–ONNX alignment & validation
├─ Converted two-stage to single-stage ONNX model → 24ms inference time reduction
└─ MMDetection3D & BEVFusion model training → 3% mAP gain
🔹 FAST-NUCES (Teaching Assistant)
└─ Graded assignments for 100+ students, mentored Python & interpolation techniques
Cricket Shot Analysis & Automated Coaching Platform
- Full-stack web app using Next.js + Flask for real-time cricket shot analysis
- Implemented CNN-GRU and Random Forest for multi-class shot classification
- Trained custom YOLO model for bat & ball detection — outperforming base YOLOv8
- Extracted joint angles via YOLOv8 Pose and compared against coaching thresholds
- Fine-tuned & quantized Gemma-2B-IT LLM to generate personalized player feedback
Logo overlay on moving people in real-time video
- Object detection + tracking pipeline using YOLOv11 + DensePose
- Extended DensePose for video with body-part segmentation & custom texture atlas generation
- Deployed CameraHMR + SMPL for 3D mesh reconstruction
- Z-buffer occlusion rendering on 6000+ vertex meshes with vectorized barycentric interpolation
- 40% temporal jitter reduction via chroma-key RGBA post-processing pipeline
Medical AI with Grad-CAM explainability
- Designed U-Net model for multi-class chest X-ray classification → 94.08% accuracy
- Pixel-level pathology segmentation — Dice coefficient 0.9818
- Applied Grad-CAM visual explainability for clinical interpretation
- Geometry-aware VTON using Meta Sapiens with normal-based deformation & lighting-aware blending
- Temporally stable video VTON for smooth, flicker-free multi-person rendering
🔽 Expand Full Project List
| Project | Tech Stack | Key Result |
|---|---|---|
| Multilingual Voice-to-SQL | Gemini 2.5 Flash, Whisper, Google TTS | 95% SQL accuracy, 75% effort reduction |
| Hybrid RAG System | LangFlow, Vector DBs, LLMs | Graph + vector retrieval, fewer hallucinations |
| AI Invoice Processing | n8n, Gemini Vision, MongoDB | 19+ fields extracted automatically |
| E-commerce Image Automation | n8n, Firecrawl, Supabase, LangChain | 40% fewer API calls |
| 3D AR Restaurant Menu | Polycam, Needle Engine, WebXR | No-app AR food visualization |
| Retail Analytics POC | DeepFace, YOLO, Streamlit | 93% precision, deployed in < 2 weeks |
| Baggage X-ray Detection | YOLOv8, RetinaNet, Faster-RCNN | 94.4% accuracy + research paper |
| Document OCR (House Maps) | AWS Textract, OpenCV | 92% accuracy, 60–70% manual effort saved |
| 3D Avatar Reconstruction | MultiPLY, PyTorch, SMPL | Monocular video → 3D avatars |
| Parkinson's Classification | Scikit-learn, SVM, Random Forest | ~90% accuracy on voice data |
| Credit Card Fraud Detection | One-Class SVM, GAN, Isolation Forest | PCA + anomaly detection |
| Vehicle Counting & Speed | YOLOv8, TensorFlow | Real-time traffic monitoring |
| Heart Disease Classification | Random Forest, Scikit-learn | 95% accuracy |
| Photo to Sketch Translation | GANs, VAEs, PyTorch | 94% accuracy + research paper |
| Incremental Face Recognition | SVM, OpenCV, Keras | 96.97% accuracy |
| Chess AI from Scratch | Python | Minimax + Alpha-Beta Pruning |
| Street Fighter II Neural Bot | TensorFlow, Keras | Key-logger trained neural network |
| 🎓 Degree | 🏫 Institution | 📅 Year | 📊 GPA |
|---|---|---|---|
| BS Computer Science | FAST-NUCES, Islamabad | June 2024 | 3.49 / 4.00 |
- 🧠 Generative AI with Large Language Models — DeepLearning.AI (2025)
- 🦜 LangChain: Chat with Your Data — DeepLearning.AI (2025)
- 🕸️ Neural Networks and Deep Learning — DeepLearning.AI (2023)
- ☁️ AWS Academy Graduate — Microservices & CI/CD Pipeline Builder (2024)
🏅 Dean's List — Awarded for 5 consecutive semesters at FAST-NUCES
🥉 Bronze Medal — 3rd position in entire batch, Semester 7
"The best way to predict the future is to build it." — Alan Kay