SINGULARITY INTERFACE · MODULE 04

Deep Learning Lab

Neural architectures that bend data like gravity bends light — from dense ANNs to convolutional networks and beyond the event horizon.

← Go to Home

EXPLORE PROJECTS

Neural Architecture Gallery

Hover or scroll over a column to reveal each project card with smooth precision.

CNN Image Classification Computer Vision

🤟 ASL Digits Recognizer

Custom 3-layer CNN achieving ~96% accuracy on ASL digits (0-9) with dual deployment — Streamlit web app + OpenCV real-time webcam inference.

Stack: Python · TensorFlow/Keras · OpenCV · Streamlit

Highlights: 3 Conv2D-MaxPool blocks | Smart Uncertainty Alerts | Webcam ROI Cropping

Live App View Code

CNN Object Detection Brand Recognition

🎯 Brand Spotter — Logo Detection

Advanced CNN-based logo detection and brand recognition system powered by deep learning for identifying and classifying company logos in images.

Stack: Python · TensorFlow/Keras · OpenCV · Deep Learning

Highlights: Multi-Brand Detection | Real-time Recognition | Transfer Learning

Live App View Code

CNN Anomaly Detection Industry 4.0

🔇 NoiseNinja — Acoustic Anomaly Detector

Unsupervised CNN Autoencoder trained on healthy machine sounds to detect acoustic anomalies in real-time via MSE reconstruction loss with precision thresholding at 0.0036.

Stack: Python · TensorFlow/Keras · Flask · Librosa · Chart.js

Highlights: Unsupervised Autoencoder | Live Mel-Spectrogram Dashboard | Real-Time Mic Streaming

Live App View Code

1D-CNN Healthcare Signal Processing

💓 Pulse Nova 1D

AI-powered detection of life-threatening heart rhythm abnormalities from raw ECG signals using a deep 1D Convolutional Neural Network.

Stack: Python · TensorFlow/Keras · Streamlit · Docker

Highlights: MIT-BIH Arrhythmia Dataset | 96.48% Accuracy | Z-Score Normalization

Live App View Code

CNN Object Detection Wildlife Tracking

🦌 FaunaFind — Autonomous Biodiversity Monitor

Production-grade, local-first ML pipeline designed to automate wildlife population tracking. Employs YOLOv8-Nano on normalized bounding boxes with a custom Streamlit analytics dashboard.

Stack: Python · YOLOv8 · OpenCV · Pandas · Streamlit

Highlights: YOLOv8 Edge Inference | Automated Quality-Control Filter | Asynchronous Data Logging

Live App View Code

CNN / U-Net Segmentation Healthcare

🧠 CORTEX-AI — Brain MRI Segmentation

Production-grade PACS-style workstation for semantic brain tumor MRI segmentation using a custom U-Net CNN trained on the MSD BraTS dataset. Features a containerized FastAPI backend with single-slice tensor execution, dropping inference latency to under 10ms.

Stack: Python · TensorFlow/Keras · OpenCV · FastAPI · Docker

Highlights: 0.835 Dice Coefficient | Live Swipe-to-Compare Viewport | Sandboxed Hugging Face Spaces Deployment

Live App View Code

👁️ CNN Projects

ANN Classification E-commerce

📱 Smart Price Predictor

AI-powered Multi-Layer Perceptron classifying smartphones into 4 price categories based on 20+ hardware specs with premium dark glassmorphism UI.

Stack: Python · TensorFlow/Keras · Streamlit · Pandas

Highlights: 4-Class Classification | 20+ Hardware Features | Confidence Scores

Live App View Code

ANN Regression Education

🎓 Student Performance Predictor

Robust ANN regression system predicting student final grades (G3) using UCI dataset, analyzing demographic, social & study factors.

Stack: Python · TensorFlow/Keras · Scikit-learn · Streamlit

Highlights: End-to-End Pipeline | Dropout & Early Stopping | Mixed Data Handling

Live App View Code

🧠 ANN Projects

LSTM Time Series Weather

🌦️ WeatherLens AI — LSTM Forecasting

End-to-end deep learning weather forecasting system using LSTM networks to predict 7-day temperature across 4 major global cities.

Stack: Python · TensorFlow/Keras · Open-Meteo API · Streamlit

Highlights: Multi-step LSTM | 4-City Coverage | Glassmorphism Dashboard

Live App View Code

LSTM Time Series Finance

🥇 Gold Price Oracle — AI Prediction

Self-updating LSTM prediction engine forecasting gold prices with live market data, autonomous retraining via GitHub Actions.

Stack: Python · TensorFlow/Keras · Streamlit · GitHub Actions

Highlights: Autonomous Self-Retraining | Live Market Feed | Smart Scaling

Live App View Code

LSTM Forecasting Agriculture

🌾 Deep Crop Yield Forecaster

Advanced Stacked LSTM based deep learning forecaster predicting agricultural crop yields using historical time series data and environmental variables.

Stack: Python · TensorFlow/Keras · Pandas · Geopandas

Highlights: Stacked LSTM Architecture | Time Series Forecasting | NDVI Integration

Live App View Code

LSTM Computer Vision Accessibility

🤟 SignSense-LSTM

Real-time sign language translation system analyzing 30 consecutive frames (temporal motion) using MediaPipe 3D Hand Landmarks and a deep LSTM network.

Stack: Python · TensorFlow/Keras · OpenCV · MediaPipe · Streamlit

Highlights: Temporal Sequence Buffer | Live Confidence Meters | Hybrid Fallback System

Live App View Code

🔄 RNN / LSTM Projects

GNN / GCN Fraud Detection Forensics SPA

🛡️ AegisGNN

Deep learning financial fraud detection system leveraging Graph Convolutional Networks (GCN) to scan transaction logs and identify high-risk networks.

Architecture: Heterogeneous user-transaction networks modeled as spatial graphs, preserving transaction amount attributes.

Optimization: 2-layer GCNConv optimized via Weighted NLL Loss (10x fraud weight) to solve severe class imbalance.

Stack: PyTorch Geometric · Flask · Vis.js · Chart.js · Python

View Code

ST-GCN GNN/GAT Time Series

🏙️ CityPulse AI

Sophisticated real-time traffic forecasting system leveraging Spatio-Temporal Graph Neural Networks (ST-GCN) to predict city-wide congestion 30 minutes into the future.

Architecture: Road networks modeled as non-Euclidean graphs using Graph Convolutions for spatial ripple effects and Temporal Convolutions.

Optimization: Huber Loss & Early Stopping | 0.55 Val Loss (transitioned from mean guessing).

Stack: PyTorch · FastAPI · Streamlit · Python

View Code

📊 GNN Projects

Transformer Fine-Tuning NLP

🧠 Emotion Pro Analytics

High-accuracy sentiment engine fine-tuning pre-trained DistilBERT on Google's GoEmotions dataset, with dual operational priorities and safety threshold gates.

Architecture: Custom fine-tuned DistilBERT classification head mapping raw feedback to 6 fine-grained emotions.

Safety: Integrated 60% confidence gate isolating ambiguous entries for human-in-the-loop review.

Stack: Python · PyTorch · Hugging Face · Streamlit · Plotly

View Code

⚡ Transformer Projects