- 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 systemstriggered 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)
🏢 L T M
📍 Bengaluru