Contribution · Application — E-commerce & Retail

Semantic Product Search for E-commerce

Keyword search fails the moment a shopper types 'shoes for marathon training' instead of 'running shoes size 10'. Semantic search embeds both the query and the catalog into the same vector space, then reranks with an LLM that can reason about intent, size constraints, and substitutions. Done right it crushes zero-result pages and long-tail conversions — done wrong it buries high-margin inventory.

Application facts

Domain
E-commerce & Retail
Subdomain
Catalog search
Example stack
Cohere Embed v4 multilingual embeddings for catalog and query · Elasticsearch or OpenSearch for hybrid BM25 + kNN retrieval · Claude Haiku 4.7 for query rewriting and intent extraction · Cohere Rerank 3 for final ordering · Algolia-style merchandising rules on top

Data & infrastructure needs

  • Multilingual product catalog with attributes and synonyms
  • Historical search logs with click and conversion signals
  • Inventory, pricing, and merchandising metadata
  • User session context — cart, locale, device

Risks & considerations

  • Loss of merchandiser control if LLM overrides business boosts
  • Latency blowing past the 200ms search budget under load
  • Hallucinated synonyms inflating irrelevant recall
  • DPDPA consent required if query logs contain personal identifiers

Frequently asked questions

Is semantic search safe for e-commerce?

Yes — it is retrieval, not generation. Risk is minimal as long as the LLM reranker is bounded to candidates from the live catalog and merchandising rules are enforced post-rerank server-side.

What model is best for semantic search?

Cohere Embed v4 and OpenAI text-embedding-3-large are the current production defaults. For reranking, Cohere Rerank 3 is the specialist; Claude Haiku 4.7 works if you need reasoning over structured attributes.

Regulatory considerations for e-commerce search?

DPDPA in India and GDPR in EU require that personalized query logs be stored with consent and retention limits. Consumer Protection (E-commerce) Rules 2020 prohibit biased display of sellers on marketplaces.

Sources

  1. Cohere Rerank 3 docs — accessed 2026-04-20
  2. MEITY DPDPA 2023 — accessed 2026-04-20

Related

  • AI Product Recommendations for E-commerce — LLM-assisted product recommendations combine embedding retrieval over catalog SKUs with session context and business rules — lifting conversion while respecting user privacy and catalog truth.
  • AI Inventory Forecasting for Retail — Inventory forecasting blends classical time-series and deep-learning models with LLM reasoning over promotions, weather, and events — reducing stockouts and overstock across thousands of SKUs.
  • AI Contract Review and Redlining — AI contract review uses LLMs to surface risky clauses, compare against playbooks, draft redlines, and negotiate against counterparty paper — freeing lawyers from first-pass review while keeping final judgment human.
  • All Applications — Browse the complete applications library
  • VSET — Programmes offered — B.Tech programs (CSE, AI&ML, AI&DS, Electronics, IIoT)