Core Assignment
- Project: User Analysis Project for TCS Innovista.
- Role: Lead Developer.
- Work Type: AI automation, backend workflows, incident analytics, infrastructure automation.
November 2024 - August 2025
Lead Developer - User Analysis Project (TCS Innovista)
Implemented an AI-powered ticket analytics platform covering ingestion, NLP processing, storage, and visualization for incident insights and automation planning.
Nov 2024 - Aug 2025
Project period
~100K / 3 months
Ticket volume handled
Up to 100 MB
Upload capacity
AI + Infra Automation
Work type
Role and Project Scope
Data Ingestion and Processing Pipeline
Data Ingestion System
- Built a web upload portal for ITSM ticket datasets, chatbot conversation logs, and knowledge-base data.
- Supported CSV and XLSX uploads up to 100 MB, including datasets near one lakh ticket records.
- Validated required schema columns and pushed accepted files to Azure Blob Storage.
Blob-Triggered Processing Architecture
- Designed ingestion-to-processing decoupling using Azure Blob Storage and Azure Functions triggers.
- Enabled scalable, asynchronous batch processing for large ticket volumes.
Preprocessing and Structuring
- Implemented ticket text cleaning, normalization, and column standardization.
- Added category mapping via dictionary lookup and prepared structured datasets for model inference.
ML and NLP Implementation
Model Pipeline
- Implemented NLP text analysis with HuggingFace Transformers.
- Integrated BART (facebook/bart-large-cnn), T5 (t5-base), and Pegasus for summarization and text transformation.
- Built the workflow on transformers with PyTorch or TensorFlow, plus Pandas, NumPy, NLTK, and spaCy.
Analytics Features Delivered
- Automated ticket categorization to identify support tower ownership.
- Implemented sentiment classification (positive, neutral, negative) to surface frustration trends.
- Added automatic summarization for long ticket descriptions to improve review speed and reporting quality.
Large-Scale Classification Impact
- Automated analysis of high-volume datasets (~100K tickets in 3 months).
- Replaced manual review cycles that typically took 2 to 3 weeks.
Automation Opportunity Detection
- Detected repeated issue patterns and candidate automation use cases from historical tickets.
- Generated insights for recurring operational bottlenecks and service-improvement planning.
Storage, Visualization, and Infrastructure
Persistence and Reporting
- Stored processed metadata and analytics outputs in Cosmos DB.
- Exposed actionable metrics in Power BI dashboards, including ticket distribution, sentiment trends, recurring categories, and automation opportunities.
Scalable Cloud Architecture
- Implemented distributed processing using Azure Blob Storage, Azure Functions, AKS, ACI, Azure App Service, and Azure Service Bus.
- Containerized ML workflows with Docker for scalable execution on AKS or ACI.
GPU and ML Runtime Setup
- Configured ML VM environments with NVIDIA GPU support, CUDA, and cuDNN for faster transformer inference.
- Provisioned development tooling with Python, Jupyter, Docker, and GitHub-based workflows.
End-to-End Workflow
Operational Flow
- Ticket CSV upload through web portal.
- Azure Blob Storage persistence and Azure Functions trigger.
- Preprocessing and NLP pipeline execution.
- Classification, sentiment, and summarization outputs.
- Cosmos DB storage and Power BI dashboard visualization.
Technology Stack
PythonHuggingFace TransformersBARTT5PegasusPyTorchTensorFlowPandasNumPyNLTKspaCyAzure Blob StorageAzure FunctionsCosmos DBAzure Kubernetes ServiceAzure Container InstancesAzure App ServiceAzure Service BusDockerGitHubPower BI