AI Projects & Research Tools

From cutting-edge research in temporal context neural networks to production-ready AI platforms. Explore my portfolio of AI innovations that bridge academic breakthroughs with real-world applications.

5+
AI Projects
16
Publications
🎯 FEATURED PROJECT - Latest Research

AttentionAI

Temporal Context Neural Networks for Microscopy

Revolutionary approach combining U-Net architecture with multi-head attention mechanisms to leverage extended temporal context in single-molecule localization microscopy. Achieves breakthrough precision by using up to 50 frames of temporal information.

🚀 Key Innovation

First application of Transformer attention mechanisms to microscopy data, enabling neural networks to correlate emitter positions across extended time sequences for unprecedented localization accuracy.

76.6%
Fits below CRLB
50x
Frame Context
PyTorch Transformers U-Net CUDA Computer Vision
View Code & Models Technical Details

Network Architecture

Input: 50 Frames (60×60px)
↓
U-Net Encoder
↓
Multi-Head Attention
↓
U-Net Decoder
↓
Localization Output

vs. State-of-the-Art

AttentionUNet
94%
DECODE
91%
ThunderSTORM
23%

Efficiency scores on CRLB dataset

More AI Projects

Research tools and production systems spanning computer vision to sports analytics

ReCSAI

Compressed Sensing + Deep Learning

Breakthrough neural network architecture combining compressed sensing with deep learning for confocal lifetime localization microscopy. Published in BMC Bioinformatics with state-of-the-art precision for irregular PSFs.

Recursive compressed sensing algorithm
Handles irregular PSFs in confocal imaging
CUDA-optimized for real-time processing
PyTorch CUDA Compressed Sensing BMC Bioinformatics
GitHub Publication

EndureXAI

Sports Analytics Platform

Production-grade AI platform for endurance sports analytics. Full-stack application demonstrating ML-powered performance prediction, training optimization, and real-time analytics for triathlon and cycling.

ML performance prediction models
Real-time training load analytics
Integration with Strava, Garmin APIs
Django TensorFlow PostgreSQL Production
Live Platform Details

LineProfiler

Biological Structure Analysis

Automated analysis tool for line-shaped biological structures with sub-pixel precision. Recognizes filaments in microscopy data and computes orientation, position, and morphological parameters for quantitative biology.

Sub-pixel accuracy measurements
Automated filament detection
Comprehensive documentation
Python OpenCV Image Analysis ReadTheDocs
Documentation Details

Impro

3D Point Cloud Renderer

High-performance OpenGL point cloud implementation with Python interface. Real-time rendering and filtering of large 3D SMLM datasets with millions of points, optimized for interactive scientific visualization.

Real-time rendering of massive datasets
OpenGL-accelerated performance
Python integration for workflows
OpenGL Python 3D Graphics Real-time
GitHub Details

AttentionAI: Technical Deep Dive

Revolutionary temporal context neural networks for microscopy

❌ The Problem

  • Emitters are active multiple times but information is wasted
  • High-density imaging requires specialized blinking buffers
  • Incompatible with live cell imaging and expansion microscopy

✅ Our Solution

  • Process up to 50 frames simultaneously with attention
  • Learn correlations across extended temporal sequences
  • Works with fluctuation-based super-resolution
  • Compatible with live imaging and expansion microscopy

🧠 Technical Innovation

U-Net Spatial Encoding

Convolutional layers extract spatial features from individual frames before temporal processing

Multi-Head Attention

Transformer attention learns temporal correlations across 50-frame sequences

Gaussian Mixture Output

Probabilistic localization with uncertainty quantification for each emitter

📊 Performance Results

Method Efficiency Score % Below CRLB Context Frames
AttentionUNet 94 76.6% 50
DECODE 91 39.9% 3
ThunderSTORM 23 17.5% 1

* CRLB = Cramér-Rao Lower Bound (theoretical minimum uncertainty). Higher percentages below CRLB indicate better precision.

🔧 Implementation

# Network Architecture
class AttentionUNet(nn.Module):
def __init__(self, hidden_dim=32):
self.unet1 = UNet(1, hidden_dim)
self.attention = MultiHeadAttention()
self.unet2 = UNet(hidden_dim, 10)
# Context: 50 frames, 60x60 pixels
# Output: probability, position, uncertainty

📈 Training Details

  • Training Data: 100,000 simulated frames with realistic photophysics
  • Context Length: 50 frames per sequence
  • Fine-tuning: Works better than training from scratch
  • Hardware: NVIDIA RTX 4090, 24-hour training
  • Validation: SMLM Fight Club contest dataset

🔬 Applications & Impact

Live Cell Imaging

Enable super-resolution without specialized blinking buffers

Expansion Microscopy

Compatible with physical sample expansion techniques

High-Density Imaging

Resolve crowded molecular environments with precision

Technology Stack

Modern tools and frameworks powering AI innovation

Python

PyTorch

TensorFlow

CUDA

OpenCV

OpenGL

Django

PostgreSQL

Docker

FastAPI

Git

GitHub

Open Source Contributions

Making AI research accessible to the scientific community

🚀 Research Reproducibility

All research projects include complete implementations, trained models, and simulation engines. This ensures other researchers can reproduce, extend, and build upon the work.

Complete source code with documentation
Pre-trained models and datasets
Simulation engines for training data
Comprehensive tutorials and examples
12+
Open Source Projects
50+
Citations of Tools

Interested in Collaboration?

Let's discuss how these AI innovations can advance your research or solve your technical challenges

Start a Project Explore Research