ecommerce_ops · ecommerce · workflow
Swiggy improves hyperlocal food search relevance using two-stage small language model fine-tuning
Swiggy's search must handle highly complex Indian food queries spanning regional dish names, dietary preferences, preparation styles, and occasion-based intents across millions of items, which simple string matching cannot resolve.
How it works
Common implementation structure
How this type of workflow is generally built, generalized across documented cases — not tied to any one vendor's stack. Click any stage to read what happens there. Specific products that implement these stages appear in “Tools commonly seen” below.
Stage 1 · Customer search query submitted
A customer submits a search query that may include dish names, dietary preferences, cuisine types, preparation styles, or occasion-based terms.
Tools used
all-mpnet-base-v2all-MiniLM-L6-v2TSDAEBM25
Outcome
The two-stage fine-tuned all-mpnet-base-v2 model outperformed all baseline variants in overall MAP and Precision@5; the system includes an incremental training pipeline and is currently in experimental phase with a small cohort.
What failed first
Generic string matching does not retrieve relevant items for complex food queries, and base pre-trained language models without domain adaptation produced inconsistent search results across hyperlocal spaces.
Results
Time savedup to 100 milliseconds
Volume996,000
Grounding & classification
Source type: technical build writeup
20 fields verified against source quotes, 2 dropped as unverifiable.
knowledge searchknowledge baseproduct catalogfailure mode describedhuman review describednamed customersource backedtools describedworkflow describedecommerceaccuracy improvementtechnical build writeupecommerce opsrag answering