Projects
Image Super Resolution through Diffusion
Read More
Code Reviewer
Read More
AI Resume Analyzer
Read More
Customer Support Application
Read More
Multimodal RAG Application
Read More
Credit Card Fraud Detection
Read More
×
Image Super Resolution through Diffusion
Key Features
Complete from-scratch implementation of the ResShift paper for efficient diffusion-based super-resolution
Residual shifting mechanism that reduces required diffusion steps to just 15 timesteps
U-Net architecture with 4-stage encoder-decoder structure and skip connections
Swin Transformer blocks in the bottleneck for global feature modeling with window-based attention
Time conditioning with sinusoidal embeddings for diffusion timesteps
Trained on DIV2K high-quality image dataset for robust performance
Comprehensive evaluation metrics including PSNR, SSIM, and LPIPS
Technologies Used
PyTorch
ResShift
Diffusion Models
Swin Transformer
U-Net
DIV2K Dataset
Computer Vision
Python
View Code
Back
×
AI Resume Analyzer
Key Features
Smart Resume Analysis powered by OpenAI's advanced language models
Precise job matching using embedding-based similarity scoring
Actionable feedback with specific improvement suggestions
Cloud-powered scalability through Azure deployment
Secure processing with enterprise-grade infrastructure
Technologies Used
OpenAI
Azure ML
Python
Embeddings
NLP
View Code
Back
×
Customer Support Application
Key Features
Advanced LLM-powered ticket categorization and response generation
Intelligent sentiment analysis for customer feedback
RAG framework implementation with FAISS vector database
Scalable deployment using Azure ML pipelines
Automated workflow with OpenAI embeddings integration
Technologies Used
Azure ML
OpenAI
FAISS
RAG
Python
LangChain
View Code
Back
×
Multimodal RAG Application
Key Features
Multimodal RAG system combining text and image processing
Restaurant app with personalized food recommendations
AWS Bedrock integration for robust processing
FAISS vector database for efficient retrieval
Claude-Sonnet AI model implementation
Technologies Used
AWS Bedrock
FAISS
Claude-Sonnet
Python
RAG
LangChain
View Code
Back
×
Credit Card Fraud Detection
Key Features
Machine learning-based fraud detection system
High accuracy transaction classification
Reduced false positives in detection
Minimized financial loss prevention
Supervised learning algorithm implementation
Technologies Used
Python
Scikit-learn
Pandas
NumPy
Machine Learning
Data Analysis
Back
×
Code Reviewer
Key Features
Distributed Training with LoRA fine-tuning on 220M parameter Microsoft model
Trained on 150k sample dataset for realistic edit suggestions
Model distillation reducing size to 80M parameters (60% cost reduction)
Containerized inference APIs for scalable deployment
ClearML integration for experiment tracking and model versioning
Technologies Used
PyTorch
LoRA
ClearML
Docker
Model Distillation
Distributed Training
View Code
Back
