DE Dost: Teaching Machines to Be a Friend on the Last Mile

Summary

DE Dost is a GenAI onboarding companion embedded in Swiggy’s delivery-executive app, designed for new DEs who receive information but lack confidence-building, immediate guidance. It supports questions in nine languages across SOPs, earnings, payouts, app navigation, shifts, and support, adapting detail and tone to conversational context. Before the first LLM call, a graph state assembles user context, preprocessing resolves references, and embedding-based semantic routing separates general knowledge-base questions from personal diagnostics. Retrieval uses authored QnA pairs plus hybrid semantic and lexical scoring to handle paraphrases and terms such as “MG” and “COD,” while guardrails constrain follow-ups to available knowledge. Evaluation spans about 800 cases per language across fleet and payout combinations, with macro accuracy trending at 90%+ and a target P50 latency of 4.5 seconds after reducing an initial 15-second pilot.

Takeaways

1. 1

   The system resolves conversational references during preprocessing, producing a normalized English query before retrieval; this prevents follow-ups such as “What if I miss it?” from being interpreted without their payout context.

2. 2

   Semantic routing uses sentence-embedding similarity against seeded general and personal utterances, avoiding an extra LLM classification call and allowing the routing model to be updated by changing and re-embedding seed examples.

3. 3

   The evaluation matrix covers about 200 questions across two fleet types and two payout types, producing roughly 800 test cases per language, with multilingual, adversarial, routing, and end-to-end response checks.