Hi, my name is
Tapos Datta.
I build things for users, powered by vision and code.
I'm a passionate and experienced Software Engineer with 7+ years of experience building high-performance mobile apps, real-time media tools, and intelligent image processing systems. With a background in computer vision and a passion for machine learning, I craft solutions that merge research, code, and user-focused design into seamless digital experiences.
About Me
Hello! I'm Tapos, a software engineer who enjoys building things that blend technology, research, and real-world impact. My journey began during my undergraduate studies in Computer Science & Engineering at Shahjalal University of Science & Technology, where I immersed myself in research, machine learning, and extensive course projects that honed both my analytical and programming skills.
During my university years, I completed numerous hands-on projects ranging from algorithm implementation to early experiments in computer vision. I also co-authored one journal paper and three conference papers, which laid the foundation for my research-oriented approach to software development.
After graduation in 2017, I joined a healthcare-focused startup where I applied my research knowledge to build real-world tools for retinal pathology detection. I developed desktop applications using C++, OpenCV, and Qt5, working on tasks such as retinal vessel segmentation, optic disc localization, and AMD stage detection.
In 2019, I moved to BrainCraft Ltd., where my career expanded into mobile development. I began in the R&D team, creating real-time image processing components for mobile using OpenGL ES and Metal, then transitioned into app development for both Android and iOS. I've since contributed as a core developer to several widely used mobile apps.
Over the years, I've continued refining my skills in native app development, computer vision, and machine learning, with a special interest in optimizing ML models for real-time and mobile environments. My current focus areas include:
- On-Device Computer Vision
- Real-Time Media Rendering
- Edge-AI Deployment
- Deep Learning for Computer Vision
- Performance-Optimized App Design
- Research-to-Product Translation
Technologies I mostly use
Languages
- Swift
- Dart
- Java
- Kotlin
- C++
- Python
- GLSL
Graphics APIs
- OpenGL ES
- Metal
- Core Graphics
- Core Image
- Core Animation
Mobile Development
- Android SDK
- Android NDK
- SwiftUI
- UIKit
- Android Jetpack
- Core Data
- Flutter
Media Frameworks
- ExoPlayer
- AVFoundation
- MediaCodec API
- Camera API
- FFmpeg
ML & Vision
- PyTorch
- OpenCV
- ONNX
- Vision API
- TensorFlow Lite
- Core ML
Tools & IDEs
- Git
- Xcode
- Android Studio
- VS Code
- Jira
- Jenkins
Professional Experience
Senior Software Engineer @ BrainCraft Ltd.
November 2021 - Present
- Lead the development of real-time media processing features for flagship mobile applications, serving millions of users
- Architected and implemented custom camera and video processing pipelines, achieving 60fps performance on mobile devices
- Spearheaded the optimization of ML models for mobile deployment, reducing model size by 40% while maintaining accuracy
- Optimized runtime memory usage, reducing peak memory footprint by 25% and improving stability on low-memory devices.
- Mentored a team of 5 junior developers, conducting code reviews and implementing best practices
- Collaborated with product teams to define technical requirements and deliver features on schedule
Software Engineer (R&D) @ BrainCraft Ltd.
July 2019 - October 2021
- Developed real-time image processing components using OpenGL ES and Metal for mobile platforms
- Created custom shaders and rendering pipelines for efficient video processing
- Implemented computer vision algorithms for real-time feature detection and tracking
- Optimized performance-critical code paths, achieving 30% improvement in processing speed
- Collaborated with the research team to port ML models to mobile devices
Software Engineer @ iHealthScreen
July 2017 - May 2019
- Developed desktop applications for retinal pathology detection using C++, OpenCV, and Qt5
- Implemented algorithms for retinal vessel segmentation and optic disc localization
- Created tools for AMD stage detection, improving diagnosis accuracy by 25%
- Collaborated with medical professionals to validate and refine computer vision algorithms
- Optimized image processing pipelines for real-time analysis of medical images
Education & Research
B.Sc. in Computer Science & Engineering
Shahjalal University of Science & Technology, Sylhet, Bangladesh
January 2013 - September 2017
Research & Publications
- Co-authored one journal paper and three conference papers in computer vision and machine learning
- Conducted research in computer vision algorithms and their practical applications
- Developed and implemented various computer vision and ML algorithms
Academic Projects
- Built hands-on projects demonstrating practical applications across web, and database systems
- Implemented core algorithms for computer vision and data processing tasks
- Developed a research-oriented mindset through academic exploration and technical depth
Leadership & Mentoring
- Collaborated with faculty members on research projects
- Mentored junior students in research and development
- Participated in technical workshops and seminars
Related Coursework
Vision & Graphics
- Computer Vision & Image Processing
- Computer Graphics & Visualization
- Digital Signal Processing
AI & Machine Learning
- Machine Learning & Pattern Recognition
- Artificial Intelligence & Neural Networks
- Deep Learning & Computer Vision
- Natural Language Processing
Core Computer Science
- Data Structures & Algorithms
- Software Engineering & Design Patterns
- Object-Oriented Programming
Networking & Systems
- Networking & Data Communication
- Operating Systems & System Programming
- Database Systems & Information Retrieval
Projects
Professional Projects
As core developer for Vintage Camera & Retro Filters, an app that brings authentic vintage photo and video effects from 1888 to the 1990s. I helped build advanced image and video processing features, including vintage filters, glitch effects, retro VHS styles, and instant film frames. My work focused on creating smooth, real-time effects that deliver a true retro aesthetic for users and content creators on platforms like TikTok.
- Core Image
- Metal
- Swift
- AVFoundation
- Computer Vision
A powerful photo and video editor with a rich suite of AI-driven features. As part of the development team, I contributed to implementing core tools such as AI-based photo enhancement, image cutout, background effects, and advanced blur filters. My work focused on integrating smart editing capabilities to improve image quality and deliver a seamless user experience.
- Core Image
- Metal
- Swift
- Core ML
- Computer Vision
An AI-powered app for removing image backgrounds with a focus on speed and precision. I developed lightweight segmentation models optimized for edge devices, capable of detecting salient objects in real time. Additionally, I implemented post-processing techniques to refine segmentation quality and enhance overall user experience.
- TensorFlow Lite
- OpenGL ES
- Android NDK
- Java
- Computer Vision
A feature-rich Android video editor that lets users add music, apply real-time filters, and create professional-quality videos. I designed and developed the full app architecture, including a real-time frame rendering pipeline using OpenGL ES and MediaCodec, with support for transitions, effects, audio mixing, and social media export.
- OpenGL ES
- MediaCodec
- Android NDK
- Java
- ExoPlayer
A sticker customization app for WhatsApp with advanced image editing tools. I contributed to developing AI-powered segmentation for background removal, implemented a lasso selection feature for freehand object isolation, and added stroke generation to outline and enhance stickers with clean, defined edges.
- Canvas
- Android NDK
- Java
- Image Segmentation
Developed desktop applications for automated retinal pathology detection using computer vision. Implemented algorithms for retinal vessel segmentation, optic disc localization, and AMD stage detection, improving diagnosis accuracy by 25%.
- C++
- OpenCV
- Qt5
- Computer Vision
- Machine Learning
Open Source Contributions
Contributed to LiTr, a lightweight hardware-accelerated video/audio transcoder for Android. Enhanced the library's capabilities by implementing custom OpenGL ES filters and integrating ExoPlayer for improved video processing. The library supports video transformation, transcoding, and frame modification with hardware acceleration. My contributions focused on expanding the filter system and improving video playback integration.
- OpenGL ES
- GLSL
- MediaCodec
- Android NDK
- ExoPlayer
- Video Processing
Contributed to Mp4Composer-android, a popular Android library for video processing using MediaCodec API. Added a new feature to enhance the library's video processing capabilities, improving its functionality for video editing and transformation. The library is widely used with over 900 stars on GitHub and supports various video operations including filtering, scaling, trimming, and transcoding.
- Java
- CameraX
- MediaCodec
- OpenGL ES
- Image Processing
Personal Projects
Developed an enhanced U²-Net pipeline for generalized retinal vessel segmentation across multiple heterogeneous datasets (DRIVE, HRF, CHASE_DB1, STARE). Implemented a learnable enhancement front-end with hybrid contrast normalization to standardize vessel visibility across different imaging conditions. Designed a patch-based training strategy to handle high-resolution images and improve domain robustness. Achieved Dice score of 0.8062 and Sensitivity of 0.8868 on a balanced cross-dataset test set, demonstrating strong generalization capabilities. Created a CoreML-based iOS demo application for real-time vessel segmentation.
- PyTorch
- U²-Net
- CoreML
- Computer Vision
- Deep Learning
- Medical Imaging
- Python
Implemented a lightweight human segmentation model using U2-Net Lite architecture trained on the P3M-10k dataset. Developed a custom data pipeline that augmented the dataset through horizontal flipping, expanding it to ~20,000 training images. Created an efficient training and validation framework with an 80:20 split ratio. The project demonstrates practical application of deep learning for real-time human segmentation tasks, achieving high accuracy in separating human subjects from complex backgrounds.
- PyTorch
- U2-Net Lite
- Computer Vision
- Deep Learning
- Data Augmentation
- Python
Developed an Android application that transforms stereo audio into immersive 7.1 surround sound. Implemented real-time audio processing to convert stereo PCM streams into multi-channel surround audio, with periodic channel distribution for enhanced spatial effects. Created a custom volume-independent visualizer for real-time audio visualization. The project includes advanced audio filtering using IIR filters for low-pass and band-pass operations in the time domain. Gained community traction with 3 stars and 4 forks on GitHub.
- Kotlin
- AudioTrack
- PCM Processing
- IIR Filters
- Custom Visualizer
- Android Audio
Developed a custom audio visualizer for ExoPlayer2 that provides high-quality PCM data visualization. Implemented direct access to raw audio data before and after processing steps, offering better quality than Android's built-in Visualizer. Created a volume-independent visualization system that maintains consistent visual feedback regardless of player volume levels. The project has gained traction with 9 stars and 3 forks on GitHub.
- Java
- ExoPlayer2
- Audio Processing
- PCM Data
- Custom Visualization
- Android Media
Developed an Android application that transforms images into artistic styles using deep learning models. Implemented real-time image processing using OpenGL ES for efficient rendering and pixel buffer operations. Integrated CartoonGAN and arbitrary style transfer models using TensorFlow Lite for on-device inference. The app uses GLSurfaceView for OpenGL ES context and efficient pixel buffer management between the view and ML models.
- OpenGL ES
- Kotlin
- TensorFlow Lite
- CartoonGAN
- Pixel Buffer
- GLSurfaceView
Integrated FFmpeg (v4.1.5) with Android NDK (r21) to create a native audio processing pipeline. Implemented efficient audio decoding and playback using AudioTrack, demonstrating practical application of native code integration in Android. The project showcases advanced audio streaming capabilities and efficient memory management between native and Java layers.
- C++
- Android NDK
- FFmpeg
- CMake
- AudioTrack
- Native Audio
Developed a Python-based tool for ML dataset preparation with interactive image cropping capabilities. Implemented a custom GUI using Tkinter for manual validation and ROI selection. Features include batch processing of image-mask pairs, keyboard shortcuts for efficient workflow, and automatic organization of cropped outputs. Designed specifically for computer vision dataset preparation.
- Python
- OpenCV
- Tkinter
- PIL
- Data Processing
- ML Tools
Built a Windows desktop application for screen recording and screenshot capture using C++ and Qt5. Implemented custom region of interest selection with OpenCV integration for screen capture. Features include multi-display support, ROI selection with mouse interaction, and flexible output options for both video recording and static screenshots.
- C++
- Qt5
- OpenCV
- Screen Capture
- Windows API
- GUI Development
Created an Android application for real-time face detection and tracking using OpenCV. Implemented efficient face detection algorithms with support for multiple face tracking and basic facial feature analysis. The app demonstrates practical application of computer vision in mobile environments with optimized performance.
- Java
- OpenCV
- Android
- Computer Vision
- Real-time Processing
- Face Detection
Developed a comprehensive pharmacy management system with features for inventory tracking, sales management, and customer records. Implemented a user-friendly interface with robust data management capabilities, demonstrating practical application of software engineering principles in healthcare management.
- Java
- MySQL
- JDBC
- Swing
- Database Design
- Desktop Application
Publications
Rating prediction for recommendation: Constructing user profiles and item characteristics using backpropagation
Applied Soft Computing
2019
Developed a novel approach for rating prediction in recommendation systems using deep learning backpropagation. The method constructs user profiles and item characteristics through separate neural networks, achieving improved prediction accuracy. Cited 14 times.
Bengali handwritten character recognition using deep convolutional neural network
20th International Conference of Computer and Information Technology (ICCIT)
2017
Presented a deep convolutional neural network approach for Bengali handwritten character recognition. The model achieved high accuracy in recognizing complex Bengali characters. Cited 90 times.
Product recommendation: A deep learning factorization method using separate learners
20th International Conference of Computer and Information Technology (ICCIT)
2017
Proposed a deep learning factorization method for product recommendation using separate learners. The approach improved recommendation accuracy through better feature learning. Cited 9 times.
Layered representation of Bengali texts in reduced dimension using deep feedforward neural network for categorization
21st International Conference of Computer and Information Technology (ICCIT)
2018
Developed a deep feedforward neural network approach for Bengali text categorization using layered representation in reduced dimensions. The method effectively handled the complexity of Bengali text processing. Cited 3 times.
Get In Touch
Let's Connect
I'm currently looking for new opportunities. Whether you have a question or just want to say hi, I'll try my best to get back to you!