Speak at The Fifth Elephant 2026 Annual Conference

Session Description

Traditional keyword search inevitably stumbles when faced with multi-lingual nuances and complex user intent. To solve this at scale, we built and launched Aslan, PhonePe’s natural language search assistant that now seamlessly handles over 1.2 million queries every single day. By breaking language barriers across English, Hindi, Hinglish, Telugu, Bengali, etc Aslan utilizes intent-based routing to bypass rigid text menus and take users directly to their intended actions.

This session pulls back the curtain on the exact three-step framework we used to build and scale this system safely and cost-effectively. We will walk through our “Accuracy First” methodology using premium models (GPT-4o) and our open-source Sentinel AI framework, detail our low-risk rollout strategy, and share how we built a defensive pipeline using Machina and semantic caching. Finally, we will demonstrate how our rigorous automated evaluation pipeline gave us the confidence to migrate to GPT-4o-mini slashing latency by 4.6x.

Key Takeaways

• The Layered Cost Funnel: Learn how to build a defensive engineering pipeline (using ML classifiers and semantic caching) to intercept noise and junk queries before they hit your LLM, drastically keeping API costs in check.
• Asynchronous Multi-Agent Architecture: Discover how to decouple user intent into specialized tools via a dynamic ToolSelectorAgent to route complex queries across parallel enterprise systems safely and accurately.

Target Audience

This session is highly beneficial for:

• Software Engineers & Architects looking for proven blueprints to deploy, evaluate, and scale multi-agent LLM frameworks in high-traffic production environments.
• Product Teams & AI Practitioners who want to transition from rigid keyword search to intent-based conversational AI without watching their API bills skyrocket.

Speaker Bio

Utsab Banerjee, Member of Technical Staff at PhonePe. Utsab is an engineer at PhonePe, where he designs high-throughput distributed systems and works on the AI charter for consumer app usecases. He led the design and optimization of Aslan, PhonePe’s multi-lingual search assistant, scaling it to over 1.2 million daily queries.