Sushant Gundla
Technical Lead - AI/ML · Bengaluru, India
I'm a Technical Lead for AI/ML and Innovation at PDI Technologies, driving agentic AI solutions, automation workflows, and ML-powered products for the convenience retail and petroleum industry. With 6+ years of experience across Generative AI, NLP, and Computer Vision, I specialize in turning complex AI research into production systems that drive real business impact. I've built agent frameworks, chatbots, RAG systems, and AI-driven products serving enterprise customers at scale. Previously at ConnectWise, I architected agentic ticket resolution systems and in-house RAG frameworks used across entire organizations. I write about building AI systems, agentic architectures, and the lessons I pick up along the way. This blog is where I think out loud.
Skills & Technologies
Experience
Bengaluru, India · Hybrid
- Leading development of Agentic AI solutions and automation workflows for PDI’s convenience retail and fuel wholesale product ecosystem
- Building MyPDI — an AI-powered intelligent hub that centralizes analytics, AI-driven insights, and operations across the retail and energy platform
- Architecting reusable frameworks for AI agents, chatbots, and RAG pipelines enabling AI-powered workflows across multiple product lines
- Implementing ML solutions for fuel pricing optimization, loyalty personalization, and customer analytics at enterprise scale
Bengaluru, India · Hybrid
- Built Agentic Ticket Resolution Recommender using the Agno framework over millions of historical records — 60% reduction in resolution time, 40% drop in support costs
- Designed and shipped an Agentic AI framework for config-driven agents deployed via REST API, supporting memory, multi-KB retrieval, tool use, and enterprise integrations
- Architected an in-house RAG framework enabling low-code RAG app deployment — adopted by all internal teams and product managers
- Built an internal chatbot for employee and client queries integrating HR, cybersecurity, and university knowledge bases — 50% reduction in support resolution time
Bengaluru, India · On-site
- Pre-trained an in-house RoBERTa-based language model on 60M job descriptions — 5%+ performance gain on downstream NLP tasks
- Led development of Curie, an LLM-powered RAG querying system over 300M job descriptions responding in under 10 seconds at production scale
- Designed Skill Clustering algorithm with 93% top-5 accuracy using Sentence Transformers semantic similarity
Bengaluru, India · On-site
- Engineered a real-time Information Retrieval system for 40+ ID card types across 13 countries using SSOD, Heatmap Regression, and OCR — 95% F1-score
- Architected end-to-end ML microservice pipelines using Django, Docker, and AKS for GPU-node deployment
- Built CNN-LSTM model achieving 92% F1-score for detecting criticality in radiology reports