Cut LLM cloud hosting costs by 92% by redesigning AI Agents to run on 8B models instead of its 235B counterpart (96% reduction in parameters).

Extended Qwen3-8B model context limits by 7X using a Hybrid Chain-of-Agents RAG architecture.

Implemented Deep Search with Qwen/Mistral for generating web-search reports of over 180k tokens.

Making AI Agent Orchestration using Chain-of-Thought system.

Built and deployed RAG-based LLMs, agentic networks, and multi-agent AI with open-source and proprietary models.

Designed LLM-driven automation generating 20K+ data rows and training 18+ ML models.

Achieved 92% accuracy in PII masking by fine-tuning BERT model for client-specific data.