Digital Asset Management with AI
Corley srl delivered an end-to-end Generative AI transformation for a European enterprise operating one of the most complex digital asset management and product information platforms in the Italian market. The client manages a catalog of tens of thousands of products and digital contents organised across a proprietary multi-level hierarchical data model with dozens of customisable attributes, product families, and variant structures. Business users had historically relied on a form-based query interface that required deep familiarity with the underlying data model — a barrier that generated significant operational overhead and limited the platform's accessibility to non-technical teams.
Corley srl designed, built, and deployed an agentic AI assistant powered by Amazon Bedrock that allows business users to query the entire catalog using natural language, in Italian or English, without needing to understand filtering logic, hierarchy structures, or API conventions. The assistant is live in a production AWS environment and has demonstrated measurable efficiency gains across catalog navigation, product discovery, and digital content retrieval workflows.
Business Challenge: Limitations of the Traditional Approach
Prior to the Generative AI transformation, catalog queries that involved multiple filtering dimensions — such as identifying products of a specific family, colour, material, and target market — required operators to manually compose multi-step form queries, navigate variant hierarchy levels, and cross-reference attribute configurations. A typical complex catalog query required an average of 10 to 12 minutes of operator time to complete accurately, including time spent consulting the data model documentation, selecting the correct product family scope, and iterating through filter refinements to reach the desired result set.
The process was not scalable: as the catalog grew in breadth and the number of active attributes increased, query complexity grew disproportionately. The client estimated that catalog navigation tasks consumed approximately 30% of the working hours of their catalog management team. Moreover, form-based search offered no conversational continuity — each query started from scratch, with no memory of previous interactions, forcing operators to reformulate context repeatedly within the same work session.
Business stakeholders identified the inability to express product discovery needs in natural language as the single largest operational friction point, and formally prioritised its resolution as a strategic initiative for the year.
Generative AI Transformation: Solution Overview
Corley srl responded to the business challenge by designing a stateful, multi-tool agentic AI assistant deployed on AWS. The solution enables end users to submit natural language requests such as "show me all black leather jackets for women from the spring collection" and receive back a structured list of matching product identifiers, accompanied by a conversational summary that mirrors the user's language and intent.
The assistant operates as an agentic workflow orchestrated by LangGraph, a stateful graph-based framework for multi-step AI reasoning. The workflow routes each user request through a sequence of specialised nodes: an intent classification node that determines whether the query targets products, digital contents, structured exports, or falls outside the assistant's operational scope; a planning node that autonomously navigates the client's data model, selects appropriate API tools, and iterates over multi-step retrieval sequences until the result set is complete; a synthesis node that composes a natural language response; and an extraction node that produces a typed JSON output for integration with downstream client interfaces.
The entire solution is deployed in an AWS production environment and is accessible at a dedicated HTTPS endpoint managed by the client's DNS infrastructure.
Quantitative Business Impact
The deployment of the Generative AI assistant produced measurable efficiency improvements across the client's catalog management workflows. The following metrics were recorded during the QA validation phase and confirmed through production usage monitoring.
Average query resolution time for complex multi-attribute product searches decreased from approximately 11 minutes per query under the form-based approach to under 45 seconds using the AI assistant, representing a reduction of over 90% in operator time per query. For straightforward single-attribute lookups, the assistant resolves queries in under 15 seconds end-to-end, compared to 3 to 4 minutes previously.
Implementation Methodology and Timeline
Corley srl structured the engagement as a five-sprint delivery programme with a duration of approximately ten weeks from kickoff to MVP delivery. The methodology front-loaded investment in measurement infrastructure before optimising the assistant's behaviour, ensuring that every subsequent change could be evaluated objectively rather than through subjective assessment.
Business Challenge: Limitations of the Traditional Approach
Prior to the Generative AI transformation, catalog queries that involved multiple filtering dimensions — such as identifying products of a specific family, colour, material, and target market — required operators to manually compose multi-step form queries, navigate variant hierarchy levels, and cross-reference attribute configurations. A typical complex catalog query required an average of 10 to 12 minutes of operator time to complete accurately, including time spent consulting the data model documentation, selecting the correct product family scope, and iterating through filter refinements to reach the desired result set.
The process was not scalable: as the catalog grew in breadth and the number of active attributes increased, query complexity grew disproportionately. The client estimated that catalog navigation tasks consumed approximately 30% of the working hours of their catalog management team. Moreover, form-based search offered no conversational continuity — each query started from scratch, with no memory of previous interactions, forcing operators to reformulate context repeatedly within the same work session.
Business stakeholders identified the inability to express product discovery needs in natural language as the single largest operational friction point, and formally prioritised its resolution as a strategic initiative for the year.
Generative AI Transformation: Solution Overview
Corley srl responded to the business challenge by designing a stateful, multi-tool agentic AI assistant deployed on AWS. The solution enables end users to submit natural language requests such as "show me all black leather jackets for women from the spring collection" and receive back a structured list of matching product identifiers, accompanied by a conversational summary that mirrors the user's language and intent.
The assistant operates as an agentic workflow orchestrated by LangGraph, a stateful graph-based framework for multi-step AI reasoning. The workflow routes each user request through a sequence of specialised nodes: an intent classification node that determines whether the query targets products, digital contents, structured exports, or falls outside the assistant's operational scope; a planning node that autonomously navigates the client's data model, selects appropriate API tools, and iterates over multi-step retrieval sequences until the result set is complete; a synthesis node that composes a natural language response; and an extraction node that produces a typed JSON output for integration with downstream client interfaces.
The entire solution is deployed in an AWS production environment and is accessible at a dedicated HTTPS endpoint managed by the client's DNS infrastructure.
Quantitative Business Impact
The deployment of the Generative AI assistant produced measurable efficiency improvements across the client's catalog management workflows. The following metrics were recorded during the QA validation phase and confirmed through production usage monitoring.
Average query resolution time for complex multi-attribute product searches decreased from approximately 11 minutes per query under the form-based approach to under 45 seconds using the AI assistant, representing a reduction of over 90% in operator time per query. For straightforward single-attribute lookups, the assistant resolves queries in under 15 seconds end-to-end, compared to 3 to 4 minutes previously.
Implementation Methodology and Timeline
Corley srl structured the engagement as a five-sprint delivery programme with a duration of approximately ten weeks from kickoff to MVP delivery. The methodology front-loaded investment in measurement infrastructure before optimising the assistant's behaviour, ensuring that every subsequent change could be evaluated objectively rather than through subjective assessment.