Hi. I'm
Surya_
I build functional AI systems from custom attention models to full stack computer vision pipelines. My work sits where machine learning intersects with end to end software engineering, driven by a strict bias for latency, clarity, and craft.
// 01 · ABOUT
Engineering first. Design aware. Curious by default.

I got into engineering the same way most people get into anything worth doing: curiosity that outgrew the tutorials. Somewhere between building small tools and breaking large ones, I realised the interesting work lives at the seam between a research idea and a product someone actually opens on a Tuesday.
Since then I've worked across deep learning, computer vision, and full stack Python often in the same project. AgroScan started as a leaf classifier and became a decision-support tool. An attendance system for a refinery had to survive lighting, latency, and zero training budget. A movie tracker turned into a study in unifying four external APIs behind one clean model.
I'm interested in intelligent systems for the long haul: perception, language, and the boring but important engineering that makes them ship. I want to work with people who care about the last ten percent as much as the first ninety.

I got into engineering the same way most people get into anything worth doing: curiosity that outgrew the tutorials. Somewhere between building small tools and breaking large ones, I realised the interesting work lives at the seam between a research idea and a product someone actually opens on a Tuesday.
Since then I've worked across deep learning, computer vision, and full stack Python often in the same project. AgroScan started as a leaf classifier and became a decision-support tool. An attendance system for a refinery had to survive lighting, latency, and zero training budget. A movie tracker turned into a study in unifying four external APIs behind one clean model.
I'm interested in intelligent systems for the long haul: perception, language, and the boring but important engineering that makes them ship. I want to work with people who care about the last ten percent as much as the first ninety.
// 02 · SELECTED WORK
Systems I've built.
A small set of the things I've shipped. Each opens a full engineering case study.
// 03 · EXPERIENCE
A timeline of the work.
Internships where I built, shipped, and learned by solving real-world engineering problems.
Nov 2025 — Jun 2026
Data Science Intern
ATEES Global · AgroScan — plant disease detection
- Designed AttCM-AlexNet in PyTorch, a custom architecture layering Channel, Spatial, and Self-Attention onto AlexNet for fine-grained leaf disease classification.
- Built the full inference pipeline in Streamlit: 224×224 preprocessing, forward pass, confidence display; all in one flow.
- Integrated a live weather REST API to generate context-aware remedy suggestions instead of static advisories.
- PyTorch
- Streamlit
- REST API
- OpenCV
- NumPy
- Python
3Attention modules224²Input resolution1APIs integratedJun 2025 — Jul 2025
Full-Stack Web Developer Intern
Grapesgenix Technical Solutions · MyMovieList — Django multi-media tracker
- Built MyMovieList end-to-end in Django + SQLite: auth, sessions, watchlists, ratings, reviews, and a custom profile dashboard.
- Integrated four external APIs simultaneously: TMDB, Jikan, Kitsu, Comic Vine; behind one unified content model for films, series, anime, and comics.
- Modelled inheritance across media types with Django ORM to keep queries clean as the catalog grew.
- Django
- Django ORM
- SQLite
- REST APIs
- HTML / CSS
- Bootstrap 5
4External APIs4Media categoriesFullAuth flowsDec 2024 — Jan 2025
Python Developer Intern
Indian Oil Corporation · Face-recognition Attendance — IOC Gujarat Refinery
- Shipped an OpenCV LBPH-based face-recognition attendance system, deployed live during Toolbox Talk sessions at the Gujarat Refinery.
- Connected the recognition pipeline to MySQL with timestamped attendance logs, eliminating manual roll calls and proxy attendance.
- Added a ReportLab export module so HR could pull PDF/CSV reports directly, and wrapped the whole thing in a Tkinter UI simple enough for zero-training use.
- Python
- OpenCV
- LBPH
- MySQL
- Tkinter
- ReportLab
→ 0Manual roll callsPDF · CSVExport formatsNoneTraining required
// 04 · STACK
Tools I reach for.
Programming
Languages I use to build.
- Python
- SQL
- C / C++
- JavaScript
AI / ML
Modelling, training, evaluation.
- PyTorch
- TensorFlow
- Scikit-Learn
- NumPy
- Matplotlib
- Seaborn
Computer Vision
Perception & image systems.
- OpenCV
- MediaPipe
- Custom CNNs
- Attention Modules
Backend
Servers, APIs, data flow.
- Django
- Django ORM
- REST APIs
- Streamlit
Frontend
Interfaces people actually use.
- HTML / CSS
- Tailwind CSS
- Bootstrap
- Tkinter
Databases
Where the data lives.
- MySQL
- SQLite
- MongoDB
- Supabase
Data Analytics
Turning data into insights.
- Pandas
- Tableau
- Power BI
Developer Tools
The engineering workbench.
- Git / GitHub
- VS Code
- WSL2
- Google Colab
- ReportLab
Currently Exploring
Expanding my engineering toolkit.
- Docker
- FastAPI
- ONNX
- Transformers