AI Services and chatbot for Grid Certification

Project Overview

As Lead AI Software Developer at CarbonFreed, I led the AI architecture and team of 4 engineers, building custom document intelligence models in Azure for grid certification automation. I trained over 20 different models for distinct form types, developed Vision Language Models for technical drawing validation, and created customer-facing chatbots. My efforts transformed manual, error-prone processes into seamless, intelligent pipelines—dramatically reducing data-entry overhead and improving user satisfaction.

Core Contributions

Dataset Preparation & Model Training

• Custom Document Models: Built custom document intelligence models in Azure for grid certification, training over 20 different models for 20 distinct form types.
• Data Processing: Processed thousands of labeled documents and created synthetic data for improved generalization, achieving excellent results across all templates and forms.
• Model Performance: Achieved high accuracy rates across all document types, significantly reducing manual validation work done by engineers.
• Validation Workflows: Built complete data extraction and validation pipeline with integration to .NET application, using Kafka for asynchronous communication between Python microservices and .NET backend.

Vision Language Models & Chatbot

• Vision Language Models: Trained Vision Language Models for validating technical drawings and schematics, ensuring accuracy of extracted component information.
• Chatbot Development: Developed customer-facing chatbot using Azure OpenAI and vector databases, implementing RAG (Retrieval-Augmented Generation) for context-aware responses based on regulatory standards.
• Thesis Supervision: Supervised master thesis student working on chatbot development and another student training custom models for component detection in electrical schematics and drawings.
• User Experience: Crafted conversational flows to guide users through common certification questions, dramatically reducing support tickets.

Messaging & Integration

• Kafka Integration: Built complete data extraction and validation pipeline with integration to .NET application, using Kafka for asynchronous communication between Python microservices and .NET backend.
• Event-Driven Architecture: Designed topic schemas for "document-extracted," "validation-completed," and "submission-ready" events, enabling scalable, decoupled pipelines.

Documentation & Operations

• Team Leadership: Led AI architecture and team of 4 engineers, coordinating development efforts and ensuring high-quality deliverables.
• Technical Documentation: Authored end-user guides and investor-facing technical briefs, detailing model performance metrics, security safeguards, and compliance checkpoints.
• CI/CD & Deployment: Wrote comprehensive unit tests, implemented GitHub Actions for CI/CD pipelines, and deployed AI microservices using Azure Functions with triggers for headless operations.
• Cloud Services & DevOps: Orchestrated deployments using Docker, Terraform, and Helm on Azure; managed storage of raw and processed documents in Azure Blob Storage.

Technologies & Tools

• Languages & Frameworks: Python, C#, JavaScript, HTML/CSS
• AI & Data Services: Azure OpenAI, Azure Document Intelligence, Vector DB
• Cloud & DevOps: Azure Functions, Azure Blobs, Docker, Terraform, Helm
• Messaging & Integration: Apache Kafka, .NET backend
• Frontend & UX: WebSockets, Flutter (for prototype dashboards)
• Collaboration & Management: GitHub, Notion, Miro

Outcomes & Impact

• Model Training Success: Trained over 20 different document models for distinct form types, processing thousands of labeled documents and synthetic data, achieving excellent results across all templates.
• Efficiency Gains: Automated data extraction and validation significantly reduced manual validation work done by engineers, accelerating certification throughput.
• Team Leadership: Successfully led AI team of 4 engineers while supervising master thesis students, demonstrating technical and mentoring capabilities.
• Complete Pipeline: Built end-to-end data extraction and validation pipeline with Kafka integration to .NET backend, enabling seamless automation.
• Advanced AI: Developed Vision Language Models for technical drawing validation and RAG-based chatbot, showcasing cutting-edge AI capabilities.