Harshad Khadilkar

Harshad Khadilkar

Lead Data Scientist

Tata Group

Harshad is a lead data scientist with the Tata Group, where his focus is on making generative AI more reliable and capable. He is also a visiting associate professor at IIT Bombay, where he teaches courses in the areas of control, optimization, and reinforcement learning. He has 12 years of experience applying intelligent algorithms to real-world applications in energy, transportation, supply chain, and finance. Harshad holds a BTech from IIT Bombay and SM and PhD degrees from the Massachusetts Institute of Technology.

Today's LLMs, and in a broader sense agentic workflows and RAG, are excellent at retrieval, summarization, and conversation. They have also been given quantitative skills by providing access to tools. However, their outputs are rarely novel or surprising. In other words, their outputs are generally boring. The focus of this talk will be on exploring ways to make the outputs more interesting. We will look at well-known approaches such as training diversity and higher temperature, but we will go on to explore ways which inject novelty more organically, through sources of directed randomness. The north star of this effort is to enable generative AI to perform effective discovery, rather than stick to the beaten path.

Read More

Managing and scaling ML workloads have never been a bigger challenge in the past. Data scientists are looking for collaboration, building, training, and re-iterating thousands of AI experiments. On the flip side ML engineers are looking for distributed training, artifact management, and automated deployment for high performance

Read More

Managing and scaling ML workloads have never been a bigger challenge in the past. Data scientists are looking for collaboration, building, training, and re-iterating thousands of AI experiments. On the flip side ML engineers are looking for distributed training, artifact management, and automated deployment for high performance

Read More
Book tickets View all speakers

We use cookies essential for this site to function well. Please click to help us improve its usefulness with additional cookies. Learn about our use of cookies in our Privacy Policy & Cookies Policy.

Show details