BigHub s.r.o.

AURES Holdings, one of the largest used car retailers in Central and Eastern Europe and operator of AAA Auto, needed to improve its mobile customer journey. Browsing thousands of cars across multiple parameters was a major pain point — especially for mobile users. To fix that, AURES teamed up with BigHub to deploy a lightning-fast AI search assistant powered by Microsoft Azure OpenAI.

Key pain points

• Complex filtering and navigation made it difficult for users to find the right vehicle.
• Mobile users struggled the most, often abandoning their search due to frustration.
• Traditional search engines returned too many irrelevant or poorly ranked results.
• Manual product tagging and keyword matching were unsustainable at scale.
• Customer experience was falling short of expectations in a highly competitive market.

‍

BigHub’s solution

BigHub developed and deployed a GenAI-powered product search assistant tailored for AAA Auto's mobile website — designed specifically to handle natural language queries and accelerate product discovery.

‍

What we delivered:

• Smart vector-based search integrated with Azure OpenAI Services (GPT model family)
• AI assistant that interprets vague, complex, or partial queries in real time
• Seamless integration with AAA Auto’s internal product database and inventory systems
• Fully mobile-optimized interface with near-instant (<1s) results

Why it works:
Unlike traditional search, the assistant understands real human queries like:

“SUV under 300,000 CZK, automatic transmission, not older than 2018” — and delivers exactly what the user wants, instantly.

‍

Technology stack:
Azure OpenAI | Vector search engine | GPT LLMs | Secure API layer

‍

Results

Engagement & business balue

• +5% increase in search result clicks, indicating better matching and user satisfaction
• 600+ users engage daily with the assistant in production
• Noticeable drop in bounce rate for mobile product search sessions

‍

Customer experience

• Instant, relevant responses lead to higher satisfaction and longer session duration
• Users find what they want faster, even with vague or misspelled inputs
• Smooth, intuitive mobile experience boosts loyalty and reduces friction

‍

Next steps

Following the successful implementation of the AI search, AURES Holdings is exploring:

• Deeper personalization based on user history and intent
• Voice search capabilities to further streamline the mobile journey
• Expansion of the assistant to handle financing, insurance, and aftersales questions

VSD, a.s., a major electricity distribution company, faced recurring losses caused by unauthorized power offtakes. Manual inspections were slow, costly, and often inefficient, with many visits leading to no findings. To address this, VSD adopted an “analytics-first” approach and partnered with BigHub to build an AI-driven fraud detection tool that analyzes consumption data and flags suspicious cases with high fraud probability.

Key pain points

• Financial losses caused by unauthorized offtakes.
• Manual, time-consuming inspections with low hit rate.
• Need to distinguish natural anomalies (e.g., seasonal consumption drops).
• Fragmented data sources and lack of unified analytical view.
• Inefficient allocation of inspection team resources.

‍

BigHub’s solution

BigHub developed an AI analytical tool capable of:

• Analyzing data from consumption meters to detect “black offtakes.”
• Reconstructing historical fraud cases to train AI models.
• Identifying similar cases based on learned behavioral patterns.
• Providing a 360° analytical view of consumption in the context of business type, turnover, ownership structure, and peer benchmarks.
• Visualizing results in a clear dashboard for faster decision-making.
• Integrating data from various internal and external sources into one platform.

This solution helped VSD transition from ex post detection to a priori detection — preventing losses before they occur.

‍

Results

• Reduced inspection costs by focusing only on high-risk cases.
• Faster anomaly detection through automated AI modelling.
• More accurate and data-driven inspections.
• Unified analytical environment for customer behavior segmentation.
• Enhanced decision-making with clear, visualized insights.

‍

Key benefits for VSD

• Optimized resource allocation for inspection teams.
• Cost reduction through efficient targeting.
• Early fraud detection and prevention.
• Single analytical platform for further development.
• Foundation for future predictive and segmentation models.

ČEZ ICT Services, part of the ČEZ Group, set out to expand the use of data analytics and artificial intelligence across its technological processes. In collaboration with BigHub, several projects were delivered demonstrating how AI and modern data infrastructure can improve operational efficiency, reduce costs, and optimize the management of critical energy assets, from heating plants to nuclear power facilities.

1. AI Platform for ČEZ ICT Services

BigHub’s role:

Delivery of the platform with reuse of existing ČEZ ICTS hardware; further support of its roadmap and applications built on top of it.

‍

Solution:

An analytical environment integrated with the central repository of technological data, enabling deployment of machine learning on large data volumes under strict performance and security requirements. The platform allows effective processing of AI workloads that were previously impossible or economically inaccessible due to technical and safety constraints.

‍

Key benefits:

• High-performance environment for AI and ML workloads.
• Reuse of ČEZ ICT Services hardware and infrastructure.
• Secure access to large volumes of operational data.
• Foundation for further AI use cases within ČEZ.

‍

2. Mělník heating plant

BigHub’s role:

Analysis of the problem, identification of the core solution, design of algorithm and methodology, assistance with result interpretation - deployment in live operation.

‍

Solution:

Reduction of load and wear of critical heating plant components through smoother power control based on continuous prediction of heat demand. The model predicts flow in the return branch of the hot-water network using historical production data and regional weather in Prague. Predictions help operators manage demand peaks more efficiently and reduce the cost of restoring stressed components.

‍

Key benefits:

• Lower mechanical stress and wear on critical equipment.
• Improved stability and control of heat production.
• Reduced maintenance and replacement costs.
• Predictive model actively used in plant operation.

‍

3. Temelín nuclear power plant

BigHub’s role:

‍Initial workshops at the plant, familiarization with complex technical context, definition of analytical approach, creation of initial models.

‍

Solution:

Prevention of unwanted vibrations and non-standard operating states of the turbogenerator using neural-network models. Although AI plays a small but important role, it helps uncover hidden correlations among dozens of technological parameters and supports a deeper understanding of system dynamics.

‍

Key benefits:

• Early detection of abnormal operation patterns.
• AI support in understanding complex process relationships.
• Safer and more stable equipment operation.

‍

Summary

BigHub supported ČEZ in building the key components for long-term adoption of artificial intelligence in the energy sector. From the shared AI Platform to predictive control and diagnostics, these projects demonstrate how AI can bring measurable value even in highly regulated, technically demanding environments.

Slow information retrieval, lengthy onboarding, and overwhelmed customer service lines. PLANEO Elektro, a leading consumer electronics retailer in the Czech Republic, tackled these challenges across dozens of its stores with the help of DIEGO, an AI assistant developed by BigHub.

Key pain points

• Store employees experienced delays when accessing operational and claims-related information.
• Customers had to wait too long for responses, which slowed down in-store operations.

• An overloaded call center led to longer customer handling times.

• Onboarding new employees was inefficient, as they struggled to find information in the company’s internal “Wikipedia.”
  ‍

BigHub’s solution

• A turnkey AI assistant, DIEGO, built on the GPT-4 language model and integrated with the client’s SharePoint system.

• Provides instant support with claims handling, internal systems, and operational processes.

• DIEGO understands natural-language queries in Czech right out of the box.

• The AI assistant is available directly in stores — no need to call the support center.

NN Insurance, a leading provider of life and accident insurance, sought to streamline and accelerate the evaluation of medical reports — a critical step in determining claim payouts. The manual process was slow, error-prone, and consumed valuable time from specialists. In collaboration with BigHub, NN deployed AI Diagnosis Mapper, an intelligent component that automates the reading, classification, and evaluation of medical reports while ensuring accuracy, transparency, and data integrity.

Key pain points

• Manual and time-intensive review of every medical report.
• Specialists had to manually link diagnoses to internal valuation tables.
• High risk of human error and limited scalability.
• Experts were burdened with routine low-value claims instead of focusing on complex cases.
• Need for integration into existing systems without disrupting internal workflows.

‍

BigHub’s solution

BigHub designed and implemented AI Diagnosis Mapper, a modular AI component based on advanced NLP models, fully integrated with NN’s internal infrastructure.

Capabilities:

• Automated reading and analysis of medical reports using language models.
• Automatic assignment of diagnostic codes and payout levels.
• Identification of complex cases that still require manual review.
• Continuous learning from historical data to improve accuracy over time.
• Seamless system integration with existing NN tools and databases.
• Full ownership and no vendor lock-in — NN retains the technology and know-how.

‍

BigHub was responsible for AI architecture, model training, integration, and performance optimization.

‍

Results

• Around 97 % accuracy in automated claim assessment.
• Claim processing time reduced from 3 days to 30 minutes in average.
• 700 diagnosis (+4000 historically).
• Increased number of claims by 10% while the team headcount decreased by 10%.
• Lower manual workload and freed capacity for specialists.
• Increased quality, transparency, and auditability of all claim decisions.
• Faster, more consistent communication with clients and improved experience.

‍

Key benefits for NN pojistovna

• Time and cost savings from automation of manual work.
• Higher service quality and consistency in decision-making.
• More efficient use of experts on complex claims.
• No vendor lock-in — full control over data, infrastructure, and future development.
• Scalability to other document types and processes within the insurance ecosystem.

‍

Next steps

BigHub and NN plan to:

• Extend the solution to other insurance workflows and document types.
• Implement anomaly-detection models for fraud prevention.
• Enhance internal analytics through integration with additional AI modules.

Direct Pojistovna, a fast-growing Czech insurer, wanted to radically accelerate and simplify claims settlement. The company faced a flood of client documents in multiple formats, from handwritten claims and invoices to photos of damage. Processing these manually was time-consuming and required specialist knowledge. To solve this, Direct Pojistovna partnered with BigHub to build an advanced agent-based system that automates document handling, evaluates claims, and enables instant payouts in straightforward cases.

Key Pain Points

• Claims handlers had to process a wide range of unstructured documents (handwritten reports, PDFs without data layers, photos, invoices, powers of attorney).
• Each case took on average 15 minutes of manual work, even for routine payouts.
• Lack of structured information made it hard to detect errors (wrong account numbers, incomplete documentation, suspicious invoice items).
• No automated system existed that could prepare a complete claim summary and support quick payout decisions.
• Processing speed was limited to working hours, leaving clients waiting.

‍

BigHub’s Solution

BigHub helped designing and delivering a modular agent-based system capable of automating claims document processing. The solution combined a pragmatic iterative development model with modern AI tools:

• Hybrid approach: A rule engine enforced business rules (e.g., “if bank account is missing, block payout”), combined with AI for attribute extraction and decision-making.
• Document Intelligence: Azure Document Intelligence extracted data from invoices, claims forms, and handwritten notes.
• Agent orchestration: LangChain and LangGraph in Python managed multiple specialized agents working together.
• Event-driven architecture: Kafka processed commands and events across the layered system.
• Cursor IDE integration: Deep use of LLMs inside the development environment enabled rapid iteration and business involvement.

‍

A unique aspect of the project was the close collaboration between BigHub and Direct’s business team. Business owner Jakub Lada was actively involved in designing agents and workflows. With LLM-supported “vibe coding,” he was able to co-create logic directly in the system, gradually evolving into a business coder role. This eliminated bottlenecks and shortened the cycle from requirement to implementation.

‍

Results

Operational efficiency

• Claims processing reduced from 15 minutes of manual work to 2 minutes after the last document is submitted.
• Fully automated processing possible in straightforward cases, without claims handler intervention.
• The system operates 24/7, unlike human claims staff limited by working hours.
• Routine cases (e.g., only missing bank account details) can now be easily flagged and prepared for quick resolution.

Accuracy and reliability

• Nearly 100% precision in invoice reading and validation.
• High success rate in evaluating claim completeness and detecting potential problems.
• Clear summaries provided to claims handlers, reducing dependence on expert knowledge.

Customer impact

• Clients with uncomplicated claims can receive payouts within minutes, improving satisfaction.
• Faster, simpler communication in the claims process has been highlighted positively by customers.

Organizational impact

• Created a new internal competency: a business coder team, led by Jakub Lada, bridging business knowledge with agile AI development.
• Shifted development practices by adopting Cursor IDE as a standard environment for LLM-powered projects.
• Freed claims handlers to focus on complex cases rather than routine paperwork.

‍

Next Steps

• Extending the agent system to other insurance product lines with similar document-heavy processes.
• Rolling out new agents to handle additional decision rules and workflows.
• Scaling the business coder approach so more non-developers in Direct can directly design and test AI agents.
• Continuing to build Direct’s internal know-how in agent-based automation with Cursor IDE as a core tool.

Datart wanted to make it easier for customers to purchase more complex products, specifically TVs, where decision-making is often complicated. In cooperation with BigHub, the company created an AI assistant that acts like an experienced salesperson – it actively asks questions, recommends suitable models, and clearly explains the differences. The result is higher conversion, more satisfied customers, and a foundation for scaling to other product categories.

Challenge

Datart aimed to simplify and improve the purchase of complex products, such as televisions, which come with many parameters, features, and marketing specifications. The goal was to build an MVP AI assistant that behaves like an experienced salesperson – it understands customer needs, asks clarifying questions, recommends suitable models, and supports decision-making.

The project focused exclusively on the TV category, where customer decision-making is often difficult and can lead to delayed purchases or even lost customers.

‍

Solution

Together with Datart, we developed an MVP version of an AI assistant for choosing TVs, which:

• Welcomes the customer, introduces itself, and explains how it can help.

• Asks the right questions – instead of clicking through dozens of filters, the customer can simply describe their needs (e.g., “I want a TV for watching sports” or “I want the biggest one for the lowest price”).

• Builds personalized recommendations from the answers, reflecting key attributes (size, picture, sound, design, price).

• Compares models and clearly explains differences in parameters – not only what is better, but also why the customer will benefit.

• Ensures a smooth transition between the AI assistant and the website – without losing context or creating frustration.

The assistant works as an independent layer – it can be launched separately from the website, in the mobile app, or directly in-store. It can be used both by customers and by sales staff.

‍

Result

The MVP project confirmed that the AI assistant:

• Increases conversion – customers who use the assistant convert at a higher rate than those who do not.

• Builds Datart’s image as an electronics specialist that truly understands products and customer needs.

• Opens a new sales channel, which can operate independently – for example, as a personal shopping advisor in-store or within the mobile app.

‍

Key benefits for Datart

• Higher customer satisfaction – thanks to 24/7 support from the AI assistant in product selection.

• Better navigation of the product range – even for customers with little technical knowledge.

• Foundation for further scaling – the project is designed to be easily expanded to other categories (mobile phones, washing machines, accessories, etc.).

• Maintained control – the assistant does not hallucinate and is connected to verified product information.

• No major e-shop changes needed – the assistant runs independently but integrates smoothly with the website.

Fortuna, a leading European betting operator, looked for ways to streamline employees’ workloads. What it lacked was a unified platform for automating routine tasks. The company therefore chose a secure, high-performance and cost-efficient answer from BigHub: an AI assistant.

Key pain points

• No unified tool to handle repetitive tasks and provide company-wide access to information.
• Employees spent excessive time on routine activities such as writing meeting minutes and searching for information.
• Security concerns had previously made the client wary of AI tools.
• Off-the-shelf AI products were unsuitable due to per-user fees and high costs.
  ‍

BigHub’s solution

• Bespoke AI assistant (LLM) running on Azure OpenAI Services with GPT-4 Turbo, fully embedded in Microsoft Teams.
• Project began with GPT-3.5 in the pilot phase and progressed through GPT-4 to GPT-4 Turbo.

• “AI colleagues” whose behaviour is tailored to specific roles (e.g., sales, HR).

• Rigorous protection of internal data (no anonymisation).

• Performance tuning of the assistant and optimisation of operating costs.
  ‍

Solution Outcomes

What the client gained through collaboration with BigHub

Business efficiency:

• AI-generated meeting minutes save 3.5 hours per meeting.
• Significantly faster information retrieval.
• Higher productivity across IT, HR 
and customer support.

Cost savings:

• 94 % lower costs compared with standard licensed solutions.

• Average running cost of the AI assistant: only USD 500 per month.

• Hidden cost drivers identified and eliminated.
  ‍

Next Steps

Our plans to continue enhancing the assistant

• Deploying a RAG system to simplify onboarding and knowledge management.

• Integrating the assistant with the internal design tool Figma.

• Enabling multimodal AI inputs—from images to documents.

Kooperativa, a major player in Central Europe’s insurance market and part of the Vienna Insurance Group, needed a way to reduce administrative strain on its sales and operations teams. Its employees were losing valuable time searching for internal information, answering routine inquiries, and handling repetitive requests. That all changed with the introduction of PROKOOP, a custom AI assistant designed by BigHub and built securely on Azure.

Key pain points

• Sales and support staff spent too much time manually searching internal systems.
• Information retrieval was inconsistent and often inaccurate.
• Administrative overhead reduced face-to-face time with clients.
• The volume of repetitive internal questions placed pressure on central methodologist and product teams.
• Kooperativa needed a secure, smart solution to support client-facing employees and optimize internal processes.

‍

BigHub’s solution

BigHub delivered a tailored enterprise AI solution built on the Microsoft Azure ecosystem, combining modern LLM agents with internal data security, infrastructure, and scalability. Two core components were implemented:

1. AI helpdesk for sales representatives
An always-on AI assistant embedded into internal systems, capable of:

• Answering queries about insurance products and claims handling
• Comparing Kooperativa’s offerings with those of competitors
• Pulling information from internal data sources and external websites
• Reducing reliance on human specialists for basic queries
• Providing 24/7 support in a secure, monitored environment

‍

2. AI Chatbot for branches
A conversational interface deployed across branch offices, helping staff:

• Resolve questions related to contract terms, claims, and offers
• Navigate complex documentation quickly using RAG (Retrieval-Augmented Generation)
• Search through structured and unstructured data in real-time
• Operate seamlessly on Azure using LangChain, LangGraph, and multiple LLMs

Technology stack:
Azure Cloud | Azure Data Lake | ADLS Gen2 | Terraform | Vector databases | Entra ID | LangChain | LangGraph | Multiple LLMs incl. OpenAI | RAG agentic framework | OCR pipeline

‍

Results

Business efficiency

• 5,000+ conversations/month across departments
• 94% of users say the assistant provides faster, more accurate information
• The AI assistant operates 24/7 and reduces response times dramatically

ROI & adoption

• Pays for itself in the first year of operation
• 44% adoption rate after just 3 months
• Clear cost-effectiveness thanks to low running costs and increased productivity

Time savings & internal impact

• 8% of managers’ time saved by reducing repetitive questions
• 1% time saved for sales reps, freeing them to focus more on clients
• Methodology and product teams significantly unburdened

‍

Subjective improvements

• Higher satisfaction among frontline employees due to easier access to information
• Improved internal knowledge sharing
• Increased consistency in responses, improving client communication

‍

Next steps

Kooperativa is now working with BigHub to expand the assistant’s capabilities further, including:

• Onboarding new employees through AI-led knowledge sharing
• Enabling voice-first interactions for even faster internal support
• Integrating multimodal inputs (e.g., documents, forms) for richer automation