Akshay Agrawal

LLM Fine-tuning: Customer Support Ticket Classification

• • Fine-tuned BERT and GPT-family models for multi-class ticket classification, automating support conversation routing at scale with high precision and recall.

• • Automated model training and deployment via Kubeflow Pipelines, implementing CI/CD for AI with full experiment tracking and lineage.

• • Optimized inference for production SLAs, integrated via REST APIs into live customer-facing microservices.

Predictive AI & GCP Model Deployment

• • Designed and customer purchase prediction Google Cloud AI Platform, transactional

• • Engineered features from large-scale MySQL datasets using Python (Pandas, NumPy); applied EDA and statistical modeling to surface key behavioral signals.

Generative AI & Agentic Document Intelligence

• • Built an end-to-end Generative AI pipeline using transformer-based LLMs for intelligent document categorization, semantic retrieval, and context-aware search across enterprise repositories.

• • Engineered agentic AI workflows with multi-step reasoning, enabling autonomous document processing aligned with MCP (Model Context Protocol) and A2A (Agent-to-Agent) interaction patterns.

• • Deployed a hybrid RAG architecture pairing FAISS vector similarity search with structured queries, improving LLM grounding and contextual retrieval accuracy.

• • Containerized model inference pipelines using Docker and deployed on Azure Kubernetes Service (AKS) for scalable, production-ready serving.