Data foundations
Scattered records become one reliable source.
AI automation · Riyadh
AI agents that carry the routine. Rules that keep it right.
ProLine builds AI agents that answer email and WhatsApp, automates workflows on a deterministic rules layer, and connects both to your own data — for logistics, retail and healthcare teams in Saudi Arabia.
Routine on the left. Decisions on the right. The seam between them is where ProLine works.
Agents take the repetitive share of the work. Rules and people keep the outcome right.
Knowledge retrieval
Answers come from your own data, with the source attached.
Deterministic rules
Fixed rules decide what must never be left to a model.
AI agents
Email and WhatsApp handled; people approve what matters.
Continue — why ProLine
Why ProLine
Smaller teams. Same workload.
Many teams now carry more work with fewer people, and quality is the first thing to slip. We move the repetitive share to AI agents and fixed rules, so people keep the judgement calls.
Brand principle
AI handles the routine. Rules keep it right. People make the call.
-
01
Answer customers faster
Agents reply on email and WhatsApp in Arabic and English, and hand anything uncertain to a person.
-
02
Rules before AI
A deterministic layer applies your policies first. The model works only inside them, and every step is logged.
-
03
Answers from your own data
Knowledge-graph retrieval connects customers, orders and records, so every answer shows where it came from.
Agent simulator
See an agent take the routine off your team.
Pick a demo and watch a short, scripted run: the agent reads the request, checks your rules, does the routine part and hands the rest to a person. A demonstration, not client data or real results.
The same four steps every time: read, check the rules, act, hand off. Your rules apply before the AI acts, and anything uncertain goes to a person. Real client examples replace these once approved for publication.
Logistics deskAgent onlineAgent is typing…
السلام عليكم، وين وصلت شحنتي رقم 4471؟
Order 4471 found in your shipping system
Rule: share status only with the registered number ✓
وعليكم السلام. غادرت شحنتك مستودع الرياض الساعة 9:40 صباحاً، والتسليم المتوقع غداً بين الساعة 2 و5 مساءً.
Can I change the delivery address?
Rule: address changes need a person
Address changes are handled by our team. I have passed your request to them with your order details, and they will reply here.
Handed to the team · order, chat and request attached
Replies in Arabic and English from approved information. Anything that changes an order goes to a person.
- Al Noor PharmacyInvoice 2291: is VAT included?Answered from approved policy
- Retail customerPlease change order 8812 to size LDraft ready · waiting approval
- Fleet clientDriver was late twice this weekComplaint · sent to a person
- PromotionsYou have won a free trip!!!Spam · archived
Draft reply · order 8812
Hello, we can change your order to size L before it ships tomorrow. Shall we go ahead?
ApproveEditApproved by the team · sent
The agent answers what your policy already covers, drafts what needs a decision, and passes complaints to a person.
Tax invoice
- Supplier
- Gulf Packaging Co.
- Invoice no.
- INV-30418
- Date
- 19 Oct 2026
- PO
- PO-7731
- Subtotal
- SAR 48,000.00
- VAT 15%
- SAR 7,200.00
- Total
- SAR 55,200.00
Extracted
- Supplier
- Gulf Packaging Co.
- Invoice
- INV-30418
- PO
- PO-7731
- Total
- SAR 55,200.00
Checks
VAT is 15% of the subtotal ✓
PO-7731 matches an open order ✓
Over SAR 50,000: needs finance approval
Saved as a draft · sent to finance with the checks
Reads invoices and forms, checks them against your rules, and leaves the approval to your team.
QuestionCEO · WhatsApp
Received
UnderstandAI agent
Plan ready
SQL database91% → 78%
Knowledge graph3 trucks down
Documents2 found
VerifyRules
Checked
AnswerWith sources
Sent
CEO · WhatsApp · Sunday 08:12
Why did on-time delivery in Jeddah drop last week?
Understood · plan
Measure: on-time deliveryRegion: JeddahPeriod: 12–18 Oct vs the week beforeAccess: the CEO may see every region ✓
Relational data
Looked up: deliveries per week in Jeddah, on time and late
91% → 78% · 47 late · 33 on routes JED-07 and JED-08
Graph · follow the links
- Routes JED-07, JED-08
- served byJeddah South depot
- trucks3 of 9 in maintenance
- waiting onBrake parts from supplier
Documents
Work order WO-552113 Oct · truck held for brake pads
Supplier email12 Oct · “shipment moved to 16 Oct”
Verify before answering
- Figures come from the database, not written by the model
- Every statement has a source
- Dates agree across the three sources
Answer to the CEO · 08:13
On-time delivery in Jeddah fell from 91% to 78% last week1. 33 of the 47 late deliveries were on routes JED-07 and JED-081, where 3 of 9 trucks were in maintenance from 13 to 16 October2, waiting for brake parts that arrived four days late3.
1Deliveries table, 5–18 Oct2Fleet graph · WO-5521 to 55233Supplier email, 12 Oct
Answers a manager’s question from your own systems: the database, a knowledge graph and your documents, searched together. Every figure links to the source it came from.
Capabilities
Four capabilities, one system.
Agents do the routine work, rules keep it right, and your own data supplies the answers.
AI Agents for Email and WhatsApp
Agents that read incoming email and WhatsApp messages, reply from approved knowledge, update your systems and pass anything uncertain to a person.
- Replies in Arabic and English
- Answers only from approved knowledge
- Hand-off to a person, with the context attached
- Every message and action logged
Typical technologies
WhatsApp Business PlatformMicrosoft 365Google WorkspaceAnthropic and OpenAI APIs
Workflow Automation with a Deterministic Layer
Fixed rules handle what must always happen the same way. AI handles the reading and judging in between, inside limits set before anything runs.
- Rules applied before the model acts
- Approval gates for high-impact steps
- Shadow running before live authority
- Audit log and rollback
Typical technologies
PythonLangGraphREST integrationsFatima Agent Harness
Relational Knowledge Graph Retrieval
Your relational data mapped as connected records, such as customers, orders, shipments or appointments, so questions are answered from the records themselves, with the source cited.
- Graph built from your existing databases
- Every answer cites its source records
- Existing access permissions respected
- Tested on your own questions before launch
Typical technologies
PostgreSQLpgvectorGraph queriesPython
Data Engineering and Analytics
Pipelines, data-quality checks and dashboards that give reporting and AI a reliable base.
- Ingestion with automated quality checks
- Dashboards built around real decisions
- Forecasting where the data supports it
- Lakehouse design on Databricks
Typical technologies
SQLPySparkDatabricksPower BIJoodei Lens
Data engineering and analytics
Clean data first. Then charts people can trust.
Real Saudi figures from public sources, drawn the way our dashboards draw them. Every chart names its source; the cleaning example uses sample records.
Data engineering
Messy records in, a report your team can trust out
Raw · from three systems
| Customer | City | Date | Amount |
|---|---|---|---|
| AHMED ali | riyadh | 21/10/2026 | 1,200 SAR |
| Ahmed Ali | Riyadh | 2026-10-21 | 1200 |
| sara m. | JED | Oct 21 2026 | SAR 350.5 |
| N/A | ? | - |
- Names trimmed and cased
- City codes spelled out
- Dates in one format
- Amounts as numbers
- 1 duplicate merged
- 1 row held for review
Clean · ready for dashboards and AI
| Customer | City | Date | Amount (SAR) |
|---|---|---|---|
| Ahmed Ali | Riyadh | 2026-10-21 | 1,200.00 |
| Sara M. | Jeddah | 2026-10-21 | 350.50 |
Held: 1 row with no customer or amount, sent to a person
Customers2
SalesSAR 1,550.50
Held for review1
Manual fixes0
Sample records for illustration · checks run the same way on every load
Logistics
Saudi port containers by type, 2023 vs 2024
−47%transshipment in 2024, while imports rose 14% and exports 9%
20232024million TEU
The total fell 11%, yet the containers Saudi businesses import and export grew. A dashboard split by type, port and lane keeps planning on the real flow.
Source: Mawani 2024 annual figures, reported by Argaam, 15 Jan 2025
Healthcare
Who pays for health care in Saudi Arabia, 2023
77.8%government, 22.2% private; the private share was 30.4% in 2016
Most spending is public. Hospitals, clinics and insurers plan beds, staff and claims better when patient, capacity and cost data are joined and clean.
Source: World Bank, World Development Indicators (WHO Global Health Expenditure Database), % of current health expenditure
Retail and customer service
People in Saudi Arabia using the internet, 2000–2024
100%since 2021, up from 2.2% in 2000
Customers already search, ask and buy online. Joining sales, stock and chat data shows which channel and branch actually sells, every morning.
Source: World Bank, World Development Indicators (ITU), % of population
Bring one report your team rebuilds by hand every week. In 30 minutes we show you what a clean, automatic version of it would look like.
Book a data reviewPlatforms and technology
Built on our own platforms.
We build with two in-house platforms and choose the rest for the job, not for the logo wall.
01 · Fatima Agent Harness
- Named agents with approval cards
- Email, documents and approved tools
- Spend caps and memory per agent
- In daily use at ProLine
02 · Joodei Lens (beta)
- Manager, engineer and analyst agents
- Cleans and profiles tabular data
- Hard cost cap per run
- Exports a pipeline that runs without AI
03 · Agents and channels
- Anthropic and OpenAI APIs
- LangGraph
- WhatsApp and email integrations
04 · Rules and knowledge
- Deterministic rules and approvals
- Relational and graph data
- Retrieval with citations
05 · Data and analytics
- Python, SQL, PySpark
- Databricks and Delta Lake
- Power BI and Tableau
Fatima and Joodei Lens are ProLine’s own platforms; Lens is in beta. Other names are text marks: ProLine claims working proficiency, not partner or certification status with any vendor.
Methodology
One governed path from task to automation.
Nothing acts on its own until it has proved itself next to your team.
- 01
Map the work
Find the repetitive share: the emails, messages and handovers that follow a pattern.
- 02
Set the rules
Write the fixed rules, approval points and limits before any model acts.
- 03
Run in shadow
The automation works alongside your team first, so results are compared before it acts alone.
- 04
Release and monitor
Hand over gradually, with logs, monitoring and a documented way to stop it.
Book a consultation
Where is your team losing time?
A 30-minute call to find the work an agent or a rule could take off your team, and what the first safe step looks like.
- 30-minute strategy discussion
- A look at one workflow or channel
- No obligation — a plain answer
- Book a strategy call through the consultation calendar.