It support · Production

Adobe enhances developer productivity with Amazon Bedrock Knowledge Bases, achieving 20% improvement in retrieval accuracy

The problem

Adobe's internal developers relied on vast wiki pages, software guidelines, and troubleshooting guides with no centralized system, making it difficult to efficiently find information for troubleshooting and software upgrades. An initial Unified Support prototype confirmed the approach but revealed scalability gaps, complex resource onboarding, poor content synchronization, and infrastructure inefficiency.

First attempt

Adobe's initial Unified Support prototype confirmed the potential of the approach but exposed scalability limitations, resource onboarding complexity, content synchronization gaps, and infrastructure inefficiency that prevented it from operating at Adobe's scale, requiring a new retrieval precision focus.

Workflow diagram · grounded in source
1
Data ingested from S3
integration
“We start by pulling data from Amazon Simple Storage Service (Amazon S3) buckets. This could be anything from resolutions to past issues or wiki pages.”
2
Documents chunked by Bedrock
ai_action
“Amazon Bedrock Knowledge Bases breaks data down into smaller pieces, or chunks, defining the specific units of information that can be retrieved. This chunking process is configurable, allowing for optimization based on the specific need…”
3
Chunks vectorized with Titan V2
ai_action
“Each chunk is passed through an embedding model (in this case, Amazon Titan V2 on Amazon Bedrock) creating a 1,024-dimension numerical vector. This vector represents the semantic meaning of the chunk, allowing for similarity searches”
4
Vectors stored in OpenSearch
integration
“These vectors are stored in the Amazon OpenSearch Serverless vector database, creating a searchable repository of information.”
5
Developer submits question
trigger
“When a user poses a question, our system competes the following steps”
6
Query auto-embedded by Retrieve API
ai_action
“With the Amazon Bedrock Knowledge Bases Retrieve API, the user's question is automatically embedded using the same embedding model used for the chunks during data ingestion.”
7
Similarity search and retrieval
ai_action
“The system retrieves the most relevant chunks in the vector database based on similarity scores to the query.”
8
Metadata filtering narrows scope
routing
“Metadata filtering empowers developers to retrieve not just semantically relevant information, but a well-defined subset of that information based on specific criteria”
9
Ranked results presented
output
“The corresponding documents are ranked based on the sematic similarity of their modest relevant chunks to the query, and the top-ranked information is presented to the user.”
Reported outcome

The Amazon Bedrock Knowledge Bases solution achieved a 20% increase in retrieval accuracy compared to Adobe's existing solution, enabled seamless document ingestion and synchronization, scaled to support thousands of Adobe developers, and contributed to reduced support costs.

Reported metrics
Retrieval accuracy improvement20%
Document retrieval accuracy vs existing solution20%
Developer support costsreduced support costs
Reported stack
Amazon Bedrock Knowledge BasesVector Engine for Amazon OpenSearch ServerlessAmazon Titan V2Ragaslangchain-aws
Source
https://aws.amazon.com/blogs/machine-learning/adobe-enhances-developer-productivity-using-amazon-bedrock-knowledge-bases?tag=soumet-20
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

The Amazon Bedrock Knowledge Bases solution achieved a 20% increase in retrieval accuracy compared to Adobe's existing solution, enabled seamless document ingestion and synchronization, scaled to support thousands of…

What tools did this team use?

Amazon Bedrock Knowledge Bases, Vector Engine for Amazon OpenSearch Serverless, Amazon Titan V2, Ragas, langchain-aws.

What results were reported?

Retrieval accuracy improvement: 20%; Document retrieval accuracy vs existing solution: 20%; Developer support costs: reduced support costs (source-reported, not independently verified).

What failed first in this deployment?

Adobe's initial Unified Support prototype confirmed the potential of the approach but exposed scalability limitations, resource onboarding complexity, content synchronization gaps, and infrastructure inefficiency that…

How is this it support AI workflow structured?

Data ingested from S3 → Documents chunked by Bedrock → Chunks vectorized with Titan V2 → Vectors stored in OpenSearch → Developer submits question → Query auto-embedded by Retrieve API → Similarity search and retrieval → Metadata filtering narrows scope → Ranked results presented.

Related it support cases