Anshuman Kumar

• Architected and led development of production-ready Multi-Agent AI systems using open-source LLMs
(Mistral 8x7B, Codestral 22B, LLavaNext 34B), delivering advanced generative conversation agents
with Retrieval-Augmented Generation (RAG) and Text-to-SQL capabilities for an OEM manufacturing
client.
• Developed and deployed scalable multimodal AI pipelines utilizing PGVector to process images, PDFs,
and unstructured documents—reducing retrieval time and improving query response accuracy for
development teams.
• Implemented MLOps pipelines with Feast-based feature stores for real-time ML inference and
automated model serving, contributing to major cost savings on supply chain projects.
• Designed, trained, and deployed XGBoost predictive models with 90%+ accuracy, leveraging Apache
Airflow for scheduled retraining, monitoring, and Spark for feature engineering data flows.
• Cross-Functional Leadership: Mentored junior engineers and collaborated with product teams to
translate business requirements into scalable AI solutions

• Cross-Functional Leadership: Mentored junior engineers and collaborated with product teams to
translate business requirements into scalable AI solutions
• Real-Time AI Applications: Built event-driven, asynchronous chat/voice applications with real-time
streaming via Twilio/Avaya APIs and Server-Sent Events (SSE) for live UI updates
• Applied AI Impact: Created COVID-19 unemployment insurance chatbot serving multiple government
organizations and Brazilian healthcare services, handling high-volume user queries during pandemic
with 99.9% uptime
• Performance Optimization: Implemented code quality standards, rigorous testing, and performance
monitoring ensuring reliable AI system deployment