DataTalks.Club Building reliable AI products in the era of Gen AI and Agents - Ranjitha Kulkarni
Oct 10, 2025
Ranjitha Kulkarni, a machine learning and NLP engineer with experience from Microsoft and Dropbox, now leads efforts at NeuBird.ai to create LLM-driven AI products. She shares fascinating insights on building reliable AI systems in the age of generative AI, emphasizing the importance of context engineering and dynamic planning. Ranjitha also discusses the evolution of agent technology, the role of retrieval in their design, and the future potential of agent marketplaces. Her practical tips on evaluating AI agents and the challenges of ensuring reliability are invaluable.
AI Snips
Chapters
Transcript
Episode notes
From Image Search To Agents
- Ranjitha described her ML journey from an undergrad image search project to CMU and Microsoft, where she worked on speech recognition for Bing and Cortana.
- She then moved to Dropbox to work on recommendations and later question-answering and early agents, which shaped her agent work today.
Joining Neubird To Automate On-Call
- Ranjitha joined Neubert.ai to build agents that can take over on-call SRE duties and reduce human wakeups at night.
- She was motivated by personal on-call pain and the startup's mission to automate incident response.
What Defines An Agent Today
- An agent autonomously completes a given task using LLMs, tools, memory and storage as needed.
- The LLM often acts as the agent's decision-making core but agents vary by orchestration and tooling.