gamma_small
SN122AI-powered product recommendations for online stores with smarter shopping suggestions
Amazon's recommendation engine drives 35% of their revenue. For millions of smaller merchants on Shopify, WooCommerce, and similar platforms, that kind of AI has always been out of reach. Bitrecs is changing that by building a decentralized recommendation engine on Bittensor, where competing AI models fight to produce the best product suggestions for online stores.
// E-commerce AI that sells for merchants.
Bitrecs is an AI-powered product recommendation system for online merchants. It runs on Bittensor as subnet 122, using a competitive network of AI models to generate personalized product suggestions for stores that previously couldn't afford anything better than basic rules-based engines.
The simple version: It's like having Amazon's recommendation algorithm working for your small online store, but instead of one company building and owning it, dozens of AI models compete to produce the best suggestions.
Centralized equivalent: Amazon Personalize, Recombee, or Shopify's built-in recommendation tools.
How it works:
- Miners submit prompt artifacts: structured configuration files specifying a model, prompt, temperature, and parameters, then make an onchain commitment to begin evaluation against rotating ecommerce benchmarks
- Validators score miner submissions based on recommendation quality, latency, diversity, coherence, and real end-user signals like add-to-cart events and conversions
- The problem it solves: Amazon attributes 35% of its revenue to recommendations. That technology is locked inside large platforms. Merchants on Shopify or WooCommerce make do with rules-based systems that don't adapt or learn.
- The opportunity: There are millions of online merchants globally without access to quality AI recommendations. A free, plug-and-play recommendation widget with actual product-market fit could reach a large portion of that market.
- The Bittensor advantage: Decentralizing the recommendation engine means no single company controls the models or the pricing. Miners compete on quality, which pushes the system toward better outputs over time without requiring any single team to maintain a state-of-the-art model.
- Traction signals: According to recent posts from community observers, Bitrecs has approximately 130 clients in its pipeline. Miners have reportedly been approaching state-of-the-art benchmark performance on the Amazon All Beauty dataset evaluated using NDCG@10.
Category: Inference and Compute | Centralized Competitor: Amazon Personalize, Recombee, Barilliance
Product recommendations are one of the most studied problems in commercial AI. Amazon's system processes billions of user interactions. The gap between what Amazon runs and what a typical Shopify merchant can access has persisted for years. Bitrecs addresses that gap directly, with a decentralized architecture that doesn't require the merchant to do anything beyond installing a plugin.
Mechanism:
According to its primary GitHub repository, Bitrecs V2 is a prompt evolution subnet. Miners don't run GPU-intensive compute. Instead, they craft and submit prompt artifacts: a structured configuration file (artifact.yaml) containing a prompt, model selection, temperature, and other generation parameters. Miners make an onchain commitment to begin evaluation.
The scoring engine employs winner-take-all (WTA) logic using epsilon-Pareto dominance to identify the best-performing miner on the frontier. Scores account for statistical robustness via epsilon tolerances, sample sizes, and linear decay factors over time, with a 3-day grace period and 5% daily reduction to a 25% floor. The top-performing miner receives the bulk of emissions. The V2 architecture separates inference from prompt evolution: the network optimizes prompt strategies and model selection rather than training models from scratch.
On the merchant side, the validated best-performing configuration powers the actual recommendation widget. Merchants receive personalized "frequently bought together" and cross-sell suggestions, with analytics to track performance.
One practical consequence of the architecture: no GPU is required to mine. Miners compete on prompt quality and model selection, lowering the barrier to entry compared to compute-heavy subnets.
The GitHub repository shows 503 commits across 2 contributors, with the most recent commit dated April 29, 2026. For a two-person team, the development pace is solid and current.
- Zero emissions currently: Bitrecs receives 0% of network emissions at the time of writing. Under Taoflow, a subnet's share of the 3,600 daily TAO is driven by net staking flows smoothed over approximately 30 days. Recent 7-day net inflow is positive at roughly 257 TAO, but the EMA hasn't turned positive yet. Sustained inflows would be needed to unlock emission share.
- Single active miner: Only one miner is active on the network. Winner-take-all scoring concentrates rewards on that one participant and reduces redundancy. This reflects either a high barrier to meaningful participation or a need for more miner outreach.
- 90-day price decline: The token has declined significantly over the past 90 days. The recent 7-day move is positive at around 13%, but the longer trend reflects sustained selloff pressure since the subnet's earlier period.
- Small development team: Two contributors on the GitHub repository. Active development pace is encouraging, but building and maintaining both the subnet infrastructure and a merchant-facing product simultaneously creates key-person risk.
- Competition: Several Bittensor subnets operate in recommendation, search, and data-driven AI categories. The specific e-commerce focus is differentiated, but broader AI infrastructure plays could overlap as the ecosystem matures.