Back to Case Study
Transforming price estimation and customer support with Agentic AI
By applying the TriStorm approach, Vstorm’s team split the project into three streams, aiming to augment and rebuild business processes with Agentic AI, incorporating fresh tech into a company with an established and strong presence on the market.
Time of work reduction
Database of client interactions
To retrieve any fact from database using Agentic Ai chatbot
The client is one of the largest engraved die manufacturers in the world, producing copper and brass sheet fed dies, rotary tools, narrow web flatbed dies, and a variety of other products.
With a heavily established position on the market and decades of experience, the company noticed the changes and transformations seizing the market, and decided to transform along with it.
Industry
Engraving
Headquarters
United States
Company size
Undisclosed
Vstorm’s impact, the TL;DR:
- Reduced pricing process from 30 minutes to few clicks
- Implemented Agentic AI database management to streamline customer service process
- Agentic AI asset management ensures that the database remains fresh and updated
The engraving market is growing dynamically and is estimated to reach $317.17 million in 2026, up from $295.38 million in 2025. In part driven by the fact that the market is undergoing a rapid technological transformation, with not only new machines and tools, but also a shift toward new customer expectations and customs.
That is why our client decided to tackle these new challenges using Agentic AI solutions.
The challenge
The client benefitted from its experience and long-standing market presence. On the other hand, this also provided a source of challenges to overcome.
- Client relationship management – their database of customer relationships and interactions had been gathered and maintained for decades, representing an asset of immense value. Yet managing it, mining it for insights, and sourcing information got increasingly difficult, especially for non-technical users.
- Pricing – price estimation in engraving is a more complicated process that needs to include multiple variables, like the material used, the complexity of content to be engraved, such as a pattern or text, and the number of engravings ordered, among others. The price had to be calculated manually by a specialist.
- Knowledge management – with every order, the clients shared specifications and details managed using enterprise-grade solutions. Before our implementation, all work had to be done manually, with unnecessary delays when something was overlooked.
- Future development – previous systems had a low level of automation and required modernisation, as well as were unsuitable for use as a foundation for new systems in the future.
Strategic alignment and planning
Cracking challenges and mapping possible use cases for the company, so those with the best ROI and cost-to-benefit ratio are picked.
Proof of Value
The team builds prototypes, experimenting with tools, models and platforms to deliver the best result while staying within the budget and timeframe.
Process augmentation
The agents are embedded into platforms depending on the use case and scenario. This is followed by extensive testing with users until knowledge and solution ownership transfer is complete.
By applying the TriStorm approach, Vstorm’s team split the project into three streams, aiming to augment and rebuild business processes with Agentic AI, incorporating fresh tech into a company with an established and strong presence on the market.
AI Agent chatbot for database management
To maximize the benefits of the database as well as make working with it easier, the Vstorm team delivered a conversational AI Agent that can answer questions about the contents of the database. To minimize the risk that, in edge cases, the agent system could return wrong results or make some mistakes, the team decided to implement a solution consisting of a main agent and a sub-agent.
The main agent handles user queries about the customers in the database, processing the vast majority of them without the need to dig deeper into the database. The Sub-agent operates using SQL queries, so that edge cases and more sophisticated data operations can be handled.
To make things even more reliable, the team decided to go for a graph approach to augment the SQL sub-agent. The logic of database queries is more about highly structured logic with clear steps to follow. By separating the agentic intelligence from the graph to-do sequence, it was possible to reduce errors and handle more complex queries.
- Graph handles: Step sequencing, state management, error propagation, parallel execution
- Agent handles: Intent analysis, query generation, result synthesis, natural language understanding
This separation dramatically improved accuracy. The SQL generation agent no longer needed to think about “what comes next,” it just focused on generating the best SQL for its specific step while the graph ensured steps were executed in the right order with validated data flowing between them.
The solution was built using the PydanticAI framework and Pydantic-graph for a graph workflow. More technical details on this process can be found in this separate case study.
Image recognition system for easier pricing
The company’s workflow for pricing included not only choosing the material, but also the need to manually transfer the client-submitted design and later calculating the final price manually.
The design is usually delivered as an image in a format popular among graphic designers. It had to be ported to the internal design and pricing platform, so an employee could account for the price. The final price consisted not only of the material used and the history of customer interactions (like a temporal or permanent discount), but also the design itself, including the need to recount pixels into centimeters.
The whole pricing process took about 30 minutes.
Vstorm delivered an image recognition system that cuts this time to only few clicks to count centimeters and deliver the estimation. It was also connected to the database management solution mentioned above, so the discounts and contextual information are visible and already included. The solution works in two modes: manual and automatic. In manual mode, the system works as before, yet leveraging the agentic AI system for database information retrieval. In automatic, the system also processes the image from the client and accounts for every step with little to no human oversight.
Effectively, the pricing time of every order was cut from 30 minutes to no more than five.
Assets management and tracking
Another app used in the system ensures that the vector database the management AI agent uses remains fresh and up to date. The company uses Microsoft Sharepoint to manage documents. The update system is operating within Sharepoint and checks if there is anything new every 15 minutes.
If anything new appears, the system updates the content of the vector database used by the database AI agent and updates the RAG that augments its operations. By doing so, it becomes unnecessary to have a technical team oversee database updates as the Agentic AI solution is basically operating “in the background.”
Building reliability
Last but not least, it is necessary to observe and quality check the system. Our Vstorm team delivered an app that checks if the user is giving the conversational AI agent negative feedback of any sort. If there is negative feedback present, the system saves the logs and stores them for the tech team to check, along with pushing a slack notification.
This approach, for example, enabled the team to spot an unexpected malfunction: sometimes the chatbot misinterpreted “die” in a “fed die” and refused to talk about vulnerable topics, along with suggesting the user seek professional help.
The effect
The solution was delivered using PydanticAI and Logfire to deliver an observable agent environment. The agent is powered by a popular cloud-based enterprise vendor.
Every component of the solution is available via API and can be intertwined with other systems, either as a supporting tool or to have its own workings augmented.
With all components in place, the company may go further and implement new solutions. One of the ideas for future incorporation is to deliver a self-service app for potential customers which provides them with an online tool to estimate the final price without waiting for an employee to run the process manually.
Meet directly with our founders and PhD AI engineers. We will demonstrate real implementations from 30+ agentic projects and show you the practical steps to integrate them into your specific workflows—no hypotheticals, just proven approaches
