Projects
Research and industry projects showcase
AI Committee
An AI-powered multi-agent evaluation system using CrewAI to assess grant applications at scale, producing scored results with structured feedback via Azure Durable Functions.
Technologies
Highlights
- Multi-agent committee with 12+ specialized AI evaluators
- Serverless fan-out/fan-in orchestration via Azure Durable Functions
- Built-in prompt injection detection and consensus validation
RAG Applications
A collection of Retrieval-Augmented Generation applications exploring vector search, knowledge graph traversal, table-aware parsing, and hybrid retrieval — all deployed on Azure Container Apps.
Technologies
Highlights
- LightRAG with hybrid vector + knowledge graph retrieval
- Table-routing RAG using Docling and Azure AI Search
- Document Intelligence RAG with hybrid keyword + vector search
Azure DevOps AI Coder
An open-source agentic AI service that automatically implements Azure DevOps Work Items by cloning the target repository, running an AI coding agent to implement the feature, and opening a Pull Request for human review.
Technologies
Highlights
- Webhook-driven automation triggered by Azure DevOps Work Item tags
- Agentic AI coding with Azure OpenAI for autonomous feature implementation
- Deployed on Azure Container Apps with Terraform infrastructure-as-code
Golden Run in Manufacturing
A system for identifying golden runs in manufacturing processes through dual framework approach for detection and condition recommendation.
Technologies
Highlights
- Dual framework architecture for detection and process recommendation
- Azure AI Services deployment
- Manufacturing process intelligence
Demand forecasting engine for consumer goods using global forecasting models to predict market demand and optimize inventory management.
Technologies
Highlights
- Global forecasting model implementation
- Deployed locally for enterprise use
- Advanced time series analysis techniques
OpenIPDM
A Desktop application with a cloud version (OpenIPDM Light) for infrastructure probabilistic deterioration modeling, providing an accessible interface for analyzing and predicting transportation infrastructure conditions.
Technologies
Highlights
Concrete Surface Cracks Detection
Web application for automated detection of cracks in concrete surfaces using computer vision and deep learning techniques.
Technologies
Highlights
Developed hierarchical reinforcement learning approaches for optimal maintenance planning of transportation infrastructure.
Technologies
Highlights
- Developed a new RL environment for infrastructure maintenance planning (InfraPlanner)
- Derived a novel hierarchical reinforcement learning approach
- Created efficient maintenance planning strategies
Created probabilistic models for estimating infrastructure deterioration using network-scale visual inspections and structural attributes.
Technologies
Highlights
- Uncertainty quantification for each inspector
- Modelling the effect of interventions
- Applied to large-scale infrastructure networks
Developed efficient geometry-based optimization approache for well placement in oil fields.
Technologies
Highlights
- Developed a new genetic operator "similarty operator" for Gentic Algorithms
- Evaluated the effectiveness of the similarity operator on synthetic and real data
- Improved convergence speedo for well placement optimization problems