Search and discovery shouldn’t feel like hunting in the dark, especially with millions of products and diverse user needs. But most eCommerce search engines still rely on outdated filters, profit-first ranking, and one-size-fits-all algorithms. Techatomic wanted to flip this script. They envisioned a hyper-personalized, unbiased AI recommendation engine that understands each user’s persona, intent, and context to deliver truly relevant results. With over 500+ merchants and a 20M+ product catalog, the goal was to ensure trust, performance, and fairness at scale.

The problem: Generic search results, profit-driven rankings, and missed intent

Despite hosting a vast catalog of products from 500+ merchants, Techatomic’s search experience struggled to meet user expectations. The results were often generic, driven by static filters or profit-oriented ranking logic that overlooked the actual intent of the user. Shoppers had to sift through pages of irrelevant options, leading to long and frustrating journeys filled with decision fatigue.  

The system failed to account for contextual cues like seasonality, user persona, or qualitative product attributes such as reviews and ratings. Without personalization or ethical ranking, users felt disconnected from the platform, and the search experience lacked trust and efficiency. Techatomic needed a smarter, context-aware solution, one that could dynamically adapt to user behavior, prioritize relevance over revenue, and scale seamlessly across millions of users and queries.

The solution: A dynamic, context-aware AI recommendation engine

GoML partnered with Techatomic to design and deploy a next-gen AI recommendation engine using Generative AI, retrieval-augmented generation (RAG), and a robust AWS-native stack. The engine balances user relevance, fairness, and real-time performance at scale.

Multi-layered personalization logic

The system maps personas using merchant preferences, categories, budgets, and previous behavior to surface recommendations tailored for "people like me."

Contextual ranking engine

Powered by LLMs (Llama 4 Maverick via AWS Bedrock), products are ranked using:

• Seasonal and occasion-based signals

• Product quality indicators (ratings, reviews)

• Past query refinements and behavioral cues

Scalable real-time infrastructure

Built using AWS Lambda, API Gateway, Step Functions, and OpenSearch to handle 20-50 queries per session for 5M+ users, while preparing for 100M+ user market growth.

The impact: Intelligent AI recommendations at scale

• 40% increase in session-to-click conversions, driven by top-7 intent-based product recommendations per query

• 25% reduction in search abandonment, thanks to unbiased, user-first ranking without ad or profit bias

• 20% faster product discovery time, enabled by context-aware filtering and real-time query optimization

• Built to scale for 5M users and 20M+ product records, with sub-second latency and support for 100M+ user market growth

Lessons for AI builders, marketplaces, and product platforms

Common mistakes to avoid

• Using ad-incentivized ranking at the cost of user trust

• Treating all users the same without persona segmentation

• Ignoring context like time, mood, or qualitative feedback

Tips for product and growth teams

• Build search around personas, not just categories

• Use RAG + vector search (OpenSearch) for dynamic discovery

• Let users guide search refinement through conversational prompts

Ready to build the future of search?

Techatomic’s journey proves that ethical, AI recommendation engines can deliver both scale and personalization.

If you're looking to build an AI recommendation engine that’s smarter, faster, and more human-centric, let GoML help bring it to life.