Ronak
Kataria

Challenge

KPIT's automotive OEM clients needed AI-driven solutions across ADAS, in-cabin intelligence, connected vehicle platforms, and the full Software Defined Vehicle (SDV) lifecycle — including on-device LLM inference on resource-constrained automotive SoCs and autonomous multi-agent AI workflows.

Solution

• Led, architected and developed an Agentic AI Enterprise Platform with multi-agent framework for the automotive SDV lifecycle — orchestrating autonomous AI workflows across development, testing, validation, and deployment stages
• Led 30+ Agentic AI interventions across automotive verticals — ADAS, in-cabin intelligence, connected vehicle platforms, and autonomous workflow automation
• Pioneered Edge AI for in-car LLM use cases — fine-tuning, quantizing (GPTQ, AWQ, GGUF), and deploying LLMs on Qualcomm SoCs for low-latency on-device inference
• Architected enterprise-scale RAG-based AI productivity tools using LangChain and LLMs (GPT-4, LLaMA, Qwen, Claude)
• Designed LLMOps accelerator platform for orchestration, fine-tuning (LoRA, PEFT), deployment, and lifecycle management
• Architected real-time conversational AI for voice-enabled in-car assistants with STT/LLM/TTS pipelines, context retention, and interruption handling
• Worked with multi-agent architectures, embedding models (Nomic, OpenAI, HF), and vector databases (Pinecone, Milvus, FAISS)

Challenge

Yes Bank required a real-time fraud detection system capable of identifying anomalous credit card transactions with high accuracy while minimizing false positives.

Solution

• Designed real-time fraud detection pipeline using Isolation Forest and XGBoost for anomaly detection on transaction streams
• Engineered features on transaction velocity, geolocation, and behavioral patterns for high-accuracy detection
• Built end-to-end MLOps infrastructure on bank's on-premise data center using Kubernetes and Kubeflow
• Implemented automated model training, serving, A/B testing, and drift monitoring pipelines

Challenge

HDFC Bank needed proactive retention strategies by predicting customer churn across retail banking segments using behavioral and transactional data.

Solution

• Designed customer churn prediction model using ensemble ML techniques (XGBoost, Random Forest, Logistic Regression)
• Applied time series behavioral analysis across retail banking segments for proactive retention strategies
• Built end-to-end MLOps pipelines on AWS with Snowflake data warehousing
• Implemented automated retraining triggers based on model performance degradation

Challenge

Enterprise clients required a unified MLOps platform that could seamlessly operate across AWS, Azure, GCP, and Databricks without vendor lock-in.

Solution

• Architected multi-cloud MLOps accelerator with abstraction layer for cloud services as part of MLOps CoE
• Built scalable LLMOps platform for RAG and LangChain workflows across GPT, LLaMA 2, T5, and Stable Diffusion
• Built NLP Document AI solution leveraging sentiment analysis, topic modeling, keyword extraction, and clustering
• Built serverless inference endpoints on AWS SageMaker with auto-scaling, A/B testing, and model versioning
• Published most-viewed technical blog on LLM operationalization with Snowflake and AWS SageMaker

Challenge

UK Fortune 500 clients needed a highly secure, scalable CRM with AI-driven automation for regulated industries including healthcare and finance.

Solution

• Architected fully serverless CRM integrating 20+ AWS services with automated deployment via CloudFormation and CodePipeline
• Led design of ML-powered Conversational AI — NLP voice/text agents, intelligent task dispatcher, real-time sentiment analysis
• Built self-learning task dispatcher that autonomously routes, prioritizes, and escalates tasks
• Developed audio language conversion and real-time closed-captioning for healthcare live-stream video
• Ensured data privacy compliance across AI features serving enterprise clients in regulated industries