MLOps.community LLM Search, UI/UX challenges, Context Engineering and the 80/20 of Eval
55 snips
Sep 5, 2025 Nishikant Dhanuka, Senior Director of AI at Prosus Group, dives into enhancing AI agents for e-commerce and productivity. He shares insights on context engineering, highlighting the importance of user history and real-time data for smarter interactions. The discussion also covers UI/UX challenges, emphasizing better search functionalities through semantic understanding. Nishikant introduces collaborative user evaluations via 'labeling parties' and stresses the need for AI to genuinely simplify lives, not complicate them.
AI Snips
Chapters
Transcript
Episode notes
Real Agents: Tukan And OLX
- Nishikant described building two agent classes: productivity tools like Tukan and e-commerce shopping assistants like OLX.
- He used these real deployments to illustrate context and search challenges at scale.
Context Beats Model Choice
- Context engineering matters more than switching between top models for real-world utility.
- Embedding real-time, messy enterprise data into prompts makes agents genuinely useful.
Pipeline Live Context Into Prompts
- Build pipelines that fetch scattered, real-time enterprise context into the prompt at request time.
- Invest engineering effort into connecting live data sources instead of relying on static DB snapshots.