DataMagic AI – Hybrid Intelligence | On-premise + cloud AI together

Architecture

Two AI agents, one intelligent platform

The DataMagic AI agent consists of two separated units. The solution runs on a provider-independent architecture, so it adapts to the client's data-protection preferences — the two agents complement each other, guaranteeing the security of company data.

🛡 DM on-premise AI

Runs on the company's internal server and accesses the entire corporate data estate through the ERP: it queries, records, navigates. Company data never leaves the corporate network. The model is trained specifically on the client's business needs.

☁ DM Cloud AI

Works with the most advanced language models available online (Anthropic Claude, Azure OpenAI, Google) and gathers web information. It obtains the needed context from external sources — market benchmarks, legal updates, document interpretation.

Logistics

Heuristic AI-assisted route planning

The DataMagic TMS module determines the optimal routes for vehicles and trips by combining advanced heuristic algorithms with artificial intelligence. We extend classic operations-research methods (Vehicle Routing Problem, Travelling Salesman Problem variants) with machine-learning models — planning also draws on the company's past trip patterns, seasonal trends and drivers' actual performance.

Multi-objective optimization – simultaneously minimizes distance travelled, fuel cost, CO₂ emissions and working time, while maximizing vehicle utilization and on-time delivery.
Complex constraint set – time windows, weight and volume limits, ADR rules, cold-chain requirements, AETR rules, plus depot opening hours and customer loading capacities.
Dynamic re-planning – on new orders arriving en route, traffic jams, weather events or breakdowns it recalculates in real time and proposes only the minimal necessary change.
Predictive travel-time estimation – more accurate arrival times based on historical GPS data, traffic patterns and time-of-day characteristics than the static data of conventional map providers.
Multimodal and multi-depot support – joint optimization of any number of sites, subcontractors and vehicle types (owned fleet, rented vehicle, carrier partner).
What-if analysis – the dispatcher can run different scenarios (opening a new depot, expanding the fleet, tariff change) and the system quantifies the expected impact.

Procurement

AI-based parts and procurement support

A truck in downtime means not just repair costs but lost freight revenue and eroded customer trust. DataMagic AI turns routine parts procurement into a strategic, data-driven process.

Predictive parts-demand forecasting – based on the vehicle's type, mileage, age, usage and repair history, it predicts which parts will fail and when (brake pads, tyres, filters, timing belt, AdBlue pump, turbo, clutch).
Smart inventory optimization – suggests minimum and maximum stock per vehicle type and site, based on supplier lead time, part criticality, purchase price and perishability.
Multi-supplier RFQ and comparison – ranks by a composite score: price, reliability, past delivery accuracy, quality-complaint rate, payment terms and delivery time.
OEM and replacement part identification – finds compatible alternatives (original, aftermarket, refurbished) by part number, VIN or text description, and compares them on price, performance and warranty.
Price monitoring and market benchmark – flags unjustified price increases and compares the price paid against the market average — creating a negotiating position.
Grouped ordering – it batches non-urgent needs for volume discounts, while ordering urgent, downtime-causing items immediately from the fastest source.
Integration with the workshop and dispatch – it coordinates servicing and procurement based on planned service dates and the AI forecast, so no vehicle stands idle waiting for a part.
Learning system – every failure, repair and supplier delivery feeds back into the model, so forecasts and supplier scores keep improving.

Documents

Automatic recognition and data extraction

Processing incoming documents, invoices, waybills and contracts is no longer manual data entry — DataMagic AI reads the content, places it into documents, and only asks for approval.

OCR in Hungarian and English – for printed and handwritten content, preserving tabular structures.
Automatic incoming-invoice recognition – seller, buyer, dates, per-line VAT rates, line items, totals and payment data go straight into the system as an incoming-invoice document.
Delivery note / waybill / CMR – sender, recipient, cargo, quantities, signature and stamp detection; automatic matching to the trip recorded in the TMS.
Contracts and certificates – key-data extraction (validity, parties, amounts, critical clauses), automatic filing into the document manager.
Feedback learning – user corrections become training data; accuracy keeps increasing during rollout.
Full audit – every recognition, confidence value and user intervention is logged.

Image analysis

Smart damage diagnostics from photos

Automatic documentation of vehicle and cargo condition — objective evidence between client and subcontractor, categorized and matched to the trip.

Automatic documentation – view classification (side, rear, load area, dashboard), licence-plate and serial-number recognition.
Damage identification – recognition of scratches, dents, paint wear and breakage; coordinate-based marking on the image with severity estimation.
Photo comparison (before / after) – automatically highlights new damage or differences based on the earlier and new image.
Cargo inspection – identification of pallet count, packaging type, hazardous-material labels and external damage.
Contextual storage – all recognized data and markings go into the document manager, with searchable metadata and in the context of the related trip / document.

AI assistant

Trained, company-specific assistant

The client gets their own assistant, based on RAG (Retrieval-Augmented Generation) technology trained on the company's context. The assistant is available on every ERP screen and understands where the user is.

What the assistant knows

The full functionality of DataMagic ERP – every module, setting, screen and process.
The client's own processes – master data, rules, the to-be operating model documented during rollout.
Current Hungarian legislation – VAT Act, Accounting Act, corporate tax, Rules of Taxation, tax-authority regulations; updated automatically.
Freight-forwarding expertise – CMR, AETR, tachograph, ADR, Incoterms, rules of road freight transport.
Warehouse logistics knowledge – WMS processes, stock-valuation methods, stocktaking rules.
Accounting and financial knowledge – posting, VAT handling, period closing, financial statements.

What the assistant does

Contextual conversation – a channel available on every screen; it understands where the user is and what they see, and gives an answer built on that context.
Launching actions in natural language – “Summarize June's empty runs by region”, “Open a new trip order with 100 pallets”, “Show the posting issue on this invoice”.
Professional advice – recommends the right VAT rate, posting entry, trip type, storage zone or accounting step — with reference to the law or internal rule.
Report and analysis creation in natural language – with instant display (table, chart, dashboard).
Onboarding and help – new users get instant answers, significantly reducing training and onboarding time.
Forecasting and anomaly detection – automatic flagging of payment delays, fuel spikes, unusual trip types and stock anomalies.

Security

Data protection, permissions, compliance

DataMagic AI is not a “black box” — it runs on the same permission, audit and GDPR model as the rest of the ERP modules.

RBAC compliance – the assistant can only access data the user starting the conversation is also authorized for. DataMagic's role-based model applies to every query.
Provider choice – cloud (Azure OpenAI, Anthropic Claude, Google) or on-premise / private cloud (e.g. Llama, Mistral) — at the client's discretion.
Data-handling guarantee – the system can be configured so that no customer data enters model training (no training on customer data); every call is traceable via a log.
GDPR-aware design – automatic data masking where needed (e.g. removing personal data before document recognition).
Full audit – every AI call, input and response, plus actions accepted / modified by the user, is logged.

Hybrid Intelligence – on-premise security and the power of the cloud, together.