Card-terminal receipt reading
The system reads a photo of a card-terminal receipt, extracts the authorisation code, amount, date and last card digits, and refuses to save if the entry does not match the receipt.
Solutions
Artificial intelligence in our systems
AI in Jusoor Tech systems: receipts, transfers, contracts and CVs read automatically with server-side checks, an executive assistant with computed figures, and governed solutions for government entities.
Artificial intelligence
We use AI where it saves staff time and prevents mistakes, not for show. This page separates what runs in our systems today from what we build for public entities within their projects.
Running in our systems today
Bank-transfer notice reading
From a photo of a transfer notice it extracts the reference, amount, date, sender and bank, and matches them to the voucher before approval.
Payment-receipt matching
It reads a payment receipt and checks its amount and date against the receipt voucher before the accountant approves it.
CV data extraction
From a PDF it extracts personal details, the photo and the passport image, and fills in the profile automatically.
Contract and ticket reading
It reads Musaned contracts and flight tickets from PDF and fills in the contract and arrival schedule instead of manual typing.
Search by description
The customer describes what they need in their own words, and the system turns it into structured search criteria and shows the best matches.
Executive assistant
A manager asks in Arabic and gets figures computed from approved measures; the model never writes a number that was not computed.
Daily executive report
The last 30 days against the 30 before, the main risks and opportunities, and an Arabic summary written automatically from the same figures.
Maps profile photo guard
It checks photos uploaded to the organisation's Google Maps profile and flags fake images and scam phone numbers.
Predictive analytics on statistical rules
Early-warning tools that run on published rules and indicators rather than an AI model, so the person in charge always knows why an alert appeared.
Unusual journal entries
Entries more than three standard deviations from the norm, reversed without reason, posted out of hours or above an authority limit.
Payment default risk
A score for each customer from their late-payment history, overdue ratio, time since last payment and amount due.
Demand forecast
The trend of upcoming demand, from a moving average with monthly seasonality, to plan resources.
Our AI capabilities for public entities
The above is working proof from systems running today. These are the capabilities we build and tailor to each entity's procedures and data.
Beneficiary service
Omnichannel virtual assistant
Answers beneficiaries in Modern Standard Arabic and local dialects on the website, app, WhatsApp and phone, and hands over to staff when needed.
An agent that completes the service
It does more than answer: it verifies identity, fills in the request, attaches the documents and tracks it until it is closed.
Intelligent voice response
A contact centre that understands the caller and serves them automatically, transferring to staff with a call summary.
Instant translation
Serving beneficiaries in their own language in chat and documents, with human review of official output.
Internal work and productivity
Process automation agents
AI agents that carry out the repetitive steps of a procedure across systems and pass anything needing a decision to staff.
Enterprise knowledge assistant
Ask questions across the entity's regulations, archive and correspondence, with the source of every answer and permissions that hide what a user may not see.
Correspondence triage and drafting
Classifying, summarising and routing incoming mail, and drafting letters and replies in the entity's style for staff to approve.
Meeting minutes
Transcribing meetings and writing the minutes, decisions and actions with owners and due dates.
Archive digitisation
Turning paper archives, manuscripts and scanned Arabic forms into searchable text and data.
Procurement, finance and compliance
Tender and bid analysis
Extracting the requirements of the terms of reference, comparing bids against them point by point, and highlighting gaps for the committee.
Contract review
Flagging missing clauses, risks and departures from approved templates before signing.
Fraud and waste detection
Models that learn the normal pattern of spending, claims and inventory, and flag anomalies before they pile up.
Continuous audit
Daily automated checks of transactions against regulations and authority limits instead of annual samples.
Data and decision support
Dashboards you can ask in Arabic
A leader asks in their own words and gets figures and charts from the entity's data, with the source of every number.
Service demand forecasting
Forecasting requests and visitor peaks by season and region to plan staff and resources.
Public opinion and rumour monitoring
Tracking what is said about the entity and its services, grouping it into issues, and giving early warning of rumours with a response workflow.
Beneficiary satisfaction analysis
Reading thousands of comments, complaints and calls to find their real causes and priorities.
Computer vision and field work
Field inspection and monitoring
Analysing photos and video from site visits to detect and document violations with place and time.
Image and document verification
Detecting edited or generated images and forged documents before they are accepted.
Crowd and facility management
Counting people and measuring crowding and waiting times from facility cameras, with identities masked.
Asset inspection
Assessing the condition of buildings and equipment from images, and scheduling maintenance before failure.
Private models and infrastructure
Arabic models inside the Kingdom
Running open-source language models on the entity's servers or a local cloud, so data never leaves.
Models tuned to the entity's data
Fine-tuning a model on the entity's terminology, regulations and forms, and measuring its accuracy before go-live.
Model operations and monitoring
Versioning and monitoring of accuracy, drift and cost, with scheduled retraining.
Custom machine learning
Classification and prediction models built on the entity's data with Python, TensorFlow and PyTorch.
How we keep AI under control
The server decides
What the model reads is compared with the entry before saving; no decision rests on the model's output alone.
Computed, not generated, figures
The model only selects and phrases; figures come from tested code, and it may not invent any number.
Failure never stops the work
If the model is down, the employee carries on manually and the entry is flagged for later review.
Logging and governance
A log of every model call, the national AI ethics principles, and personal data protection.
AI FAQ
Is the entity's data sent to models outside Saudi Arabia?
Only with the entity's approval. We run the model on the infrastructure it chooses, including open-source models on servers inside the Kingdom, and we mask personal data before any external processing.
What if the model misreads a document?
Before saving, the server compares what the model read with what the employee entered. If they differ, the transaction is rejected and logged. If the document cannot be read, the manual entry is accepted and marked "not verified automatically".
Does the assistant write database queries?
No. It only picks a measure and a period from an approved list; the figures are computed by tested code, and the model is not allowed to write any number that was not computed.
Request a use-case study for your entity
Tell us which procedure takes up your team's time, and we will come back with a proposed solution and timeline.