I work on computer vision that holds up outside the lab: on crowded roads, in classrooms and in humanitarian settings.
I give what I learn away for free.
I am looking for a fully funded PhD in AI and for sponsors. Every sponsored project is released open for all.

• 🎯 Seeking: a fully funded PhD in AI, in computer vision and deep learning.
• 🔬 Research focus: robust, efficient and fair vision models that work outside benchmark conditions, for road safety, humanitarian settings and education.
• 📝 Research record:
  • 1 paper in the publication process: real-time facial-expression analysis with YOLOv11.
  • 3 undergraduate theses: pedestrian risk, vehicle over-speed and brain-tumor classification.
• 💼 Now: Advanced ICT Officer at CNRS-UNHCR since October 2025. I teach students digital and ICT skills in a humanitarian setting.
• 🧪 Before: Research Assistant at Robo Tech Valley, Dhaka (March–July 2025), working on AI/ML and computer-vision projects.
• 🏛️ Education: B.Sc. in Computer Science & Engineering, Ahsanullah University of Science and Technology (AUST), 2025.
• 🌍 Open science:
  • 273 free written lectures and 110 free animated video lectures;
  • every sponsored project is released open for all.
• ⚡ Algorithms: 10+ years of competitive programming.

My research interests centre on computer vision and deep learning for real-world safety and human understanding, particularly in developing countries such as Bangladesh.

My undergraduate research tackled pedestrian detection with risk-level classification for Bangladeshi road scenes, and vision-based vehicle over-speed detection. My current work studies real-time facial-expression analysis with YOLOv11.

I want to develop robust, efficient and fair vision models that work reliably outside benchmark conditions, for road safety, humanitarian applications and education.

1. Beyond the benchmark: How can detectors stay reliable on dense, unstructured roads, with rickshaws, buses and pedestrians sharing a lane in rain or at night, when they were trained on data from elsewhere?
2. Real-time on modest hardware: How far can YOLO-class models be compressed for edge devices without losing accuracy on the cases that matter most for safety?
3. Fair human understanding: Do expression and pose models work equally well across ages, skin tones and cultures? How do we measure the gaps and close them?
4. AI for humanitarian settings: How can vision and learning systems responsibly support education and services in refugee and low-resource contexts?

I build in the open, but compute, data and publishing cost real money.

My pledge: when a project is sponsored, I release it openly and free for anyone to learn from, use and build on. That includes:

• the source code;
• the trained model weights;
• the datasets, with personal data anonymised and consent respected;
• an open-access paper;
• lessons on my free teaching sites.

What sponsorship pays for

• 🖥️ GPU compute to train and evaluate vision models
• 🏷️ Collecting and annotating real-world data, such as Bangladeshi road scenes
• 📄 Open-access publication fees, so the papers are free to read
• 🎬 More free lectures and animated lessons
• ✈️ Travel to present the work at conferences

Projects open for sponsorship

Sponsors get:

• acknowledgement in the repository, papers and lectures;
• regular progress updates;
• a direct line to me throughout the project.

• The AI Lecture Hall · source
  • 273 university-level lectures in 10 tracks.
  • Covers AI foundations and mathematics through transformers, diffusion models, reinforcement learning, MLOps and AI ethics.
  • Each lecture includes derivations, code and exercises.
• AI in Motion · source
  • 110 animated video lectures: more than 8 hours, including 36 deep dives of 10+ minutes.
  • Narration, subtitles, chapters, quizzes and printable illustrated notes.
  • Every animation is computed live from the real algorithm: gradient descent really descends and Q-learning really learns.
• Teaching at CNRS-UNHCR: ICT and digital skills for students.
• YouTube: Study with Janin and Pomodoro Study with Janin.

• 10+ years of contest problem solving, since 2014, on Codeforces, CodeChef, HackerRank and other platforms.
• Recognised for outstanding performance in online programming contests.
• 📈 Profile: StopStalk
• 📂 Archives:
  • 10 years of competitive-programming resources
  • Codeforces solutions

Area	Tools
AI and machine learning
Algorithms and systems
Web and backend
Data
Tools

• Advanced Learning Algorithms · DeepLearning.AI & Stanford Online
• Unsupervised Learning, Recommenders, Reinforcement Learning · DeepLearning.AI & Stanford Online
• Divide and Conquer, Sorting and Searching, and Randomized Algorithms · Stanford University
• Advanced Algorithms and Complexity · UC San Diego
• Object-Oriented Data Structures in C++ · University of Illinois Urbana-Champaign
• Data Scientist Career Track · 365 Data Science: exams in ML, deep learning, SQL, mathematics, probability and statistics
• Convolutional Neural Networks with TensorFlow in Python · 365 Data Science
• MLOps Bootcamp · DataMites
• Introduction to DevOps · IBM

📁 All 39 certificates (Google Drive)

• 🥇 Talent Pool Scholarship (2008): ranked 4th in Kurigram District.
• 🏆 Competitive-programming awards for performance in online contests.
• 🧠 Microsoft MVP Program trainee, Dhaka: workshop on AI, computer vision and Microsoft Azure.
• 🔶 Google Developer Group TensorFlow workshop, Dhaka: AI, machine learning and data mining.

• 📧 Email: japurba97@gmail.com
• 🌐 Academic portfolio: janin-a-apurba.vercel.app
• 📄 CV: Google Drive
• 👥 References: academic references from AUST CSE faculty are available on request.

If my work helped you, please follow and subscribe:

• Study with Janin: youtube.com/@studywithjanin
• Pomodoro Study with Janin: youtube.com/@pomodorostudywithjanin3326

_{© Janin A Apurba, CSE, AUST · Advanced ICT Officer, CNRS-UNHCR}