• GenAI Agentic AI Strategy
• Select and optimize opensource LLMs Llama 3 Mistral Falcon BLOOM for local deployment lead finetuning using PEFTLoRAQLoRA and full finetuning strategies
• Architect LLM distillation pipelines to create smaller domain specific models for inference efficiency
• Design and govern RAG pipelines standard RAG Graph RAG Hybrid RAG for chatbots grievance redressal helpdesk automation and document summarization
• Build reusable AI Agent frameworks task agents reasoning agents retrieval agents workflow managers using Lang Graph and Auto Gen
• Architect multiagent orchestration systems with supervisor worker patterns tool use agents and self reflection loops
• Classical ML Advanced Analytics
• Design ML pipelines for calibration models ensemble models stacking blending boosting uplift causal ML models and simulation models
• Establish XAI Explainable AI frameworks using SHAP LIME Integrated Gradients define interpretability standards for regulatory compliance
• LLM Evaluation Quality Frameworks
• Create accuracy and evaluation frameworks for summarization ROUGE BERTScore LLMasjudge extraction classification and conversational AI
• Implement hallucination detection factuality scoring and bias evaluation frameworks using LangKit GuardrailsAI and custom evaluators
• Define LLM evaluation benchmarks aligned to use cases establish continuous evaluation pipelines triggered post finetuning




• Govern prompt engineering standards fewshot template libraries and chainofthought reasoning frameworks
• Document Intelligence Knowledge Pipelines
• Build document intelligence workflows OCR Tesseract PaddleOCR parsing chunking strategies embeddings summarization extraction QA
• Design semantic search and hybrid search dense sparse architectures using vector databases PGVector Milvus Qdrant Elasticsearch
• Pipeline Automation MLOps LLMOps
• Design CICD pipelines for model retraining for LLM RAG and multiagent systems triggered automatically on data driftconcept drift metrics
• Implement automated AB testing and shadow deployment frameworks for controlled model rollouts
• Define SLAs for model inference latency throughput and availability implement performance dashboards
• Security Compliance Governance
• Ensure all model artifacts weights and dependencies are managed via a local secure repository with no external API calls
• Implement PII redaction layers data anonymization and differential privacy measures in all pipelines
• Define AI governance policies model cards data lineage audit trails and biasfairness monitoring
• Collaborate with domain teams DevOps and security teams to ensure regulatory compliance
• Design observability stack for MLLLM systems latency throughput GPU utilization drift s cost tracking
• Architect feature stores and data for reproducible ML experiments

📌 Gen AI Architect (Bengaluru)
🏢 LTM
📍 Bengaluru