PC Parts sellers build listings 3x faster with an AI agent on Amazon Bedrock

Second-hand PC parts are messy to catalog: inconsistent titles, model numbers buried in descriptions, and non-technical sellers. A buyer who types "quiet 1440p build for around 3000 lei" gets nothing from a plain keyword box, and a seller who cannot describe their hardware correctly never completes a good listing.

For an early-stage marketplace, both failures kill the transaction before it starts: search that returns nothing loses the buyer, and listings that are hard to create lose the seller. PC Parts needed to lower both barriers without hiring a support team to do it manually.

We built three AI capabilities on Amazon Bedrock, using Bedrock Inline Agents with Claude as the reasoning engine and Lambda functions as the agent tools. The agents do not answer from the model's general knowledge: they call tools that read and write the live marketplace catalog in Amazon DynamoDB.

1. Intent-based search: a buyer describes what they want in natural language; a Bedrock Inline Agent extracts structured criteria and calls Lambda tools that run a structured query over the DynamoDB catalog, returning matching listings. A model cascade (Claude Haiku, then Claude Sonnet, then a non-AI keyword fallback) keeps search answering even under load or a failed call.
2. Conversational listing creation: the listing agent collects specifications through conversation and analyzes uploaded photos to identify the hardware, then auto-structures the listing with the correct category and specs, so a non-technical seller can post something buyers can actually find.
3. The "Rig" assistant answers hardware-compatibility and buying questions, grounded two ways: a system prompt carrying the marketplace category taxonomy and rules, and live tool calls against the catalog, rather than guessing from general knowledge.
4. Amazon Bedrock Guardrails and the non-AI fallbacks keep the experience safe and available, while EventBridge, S3, CloudFront, and SES support moderation, image storage, delivery, and email around the AI core, all running against the Flask application on Amazon ECS Fargate.

Sellers now build a complete, correctly-categorized listing about 3x faster than filling the form by hand, with the agent identifying components from photos and structuring the specs.

Buyers reach relevant listings by describing what they want instead of guessing keywords, and routine compatibility questions are answered by the assistant rather than the founding team.

The AI features run on the same ECS and Bedrock stack as the marketplace itself, so new capabilities ship without standing up separate infrastructure.