Applied AI
AI applications for the work your team repeats, built on your knowledge, data and systems.
We build applications that read what arrives, apply your rules and product knowledge, work with the systems you already use, and prepare the result for your team to review: a quotation, a complete application, a record ready for a decision.
What you can commission. An application your team uses every day.
One recurring task, done end to end
An application built around a task your team handles every day, from what arrives to what goes out.
Built on how your business works
Your product knowledge, your rules and policies, and the systems your team already uses.
Your team stays in charge
The application prepares the work. Your people review, adjust and approve before anything leaves.
Proven on your own cases first
Before it goes live, we test it on a sample of your real work and agree what good looks like.
For sales teams at specialist retailers and distributors
Turn product enquiries into quotations
Your team spends its time on the customer, not on searching catalogues and building the quote.
What arrives
A customer enquiry by email, web form or chat, in the customer's own words and often missing details.
What the application does
Reads the requirements, works out what's needed, finds products that fit, checks stock and price, and lists what it still needs to know.
What your team receives
Ranked options with the reasons each one fits, a draft quotation, and the questions to confirm with the customer.
What happens next
A salesperson reviews, adjusts and approves. The quotation goes out from your existing quoting system.
Quotation workspace
Sample workspace · Highveld Power Supplies
Customer enquiry
From Pieter van Wyk, Van Wyk & Mokoena Attorneys · by email
Hi there, We keep losing work during load-shedding. We need backup power for our office: six desktop computers, the Wi-Fi router and network switch, the office lights and the kitchen fridge. The laser printer can stay off. We'd want about 4 hours of power. Budget is around R90,000 including installation. We're in Pretoria East. Can you send us a quote? Pieter van Wyk Van Wyk & Mokoena Attorneys
Requirements read from the enquiry
Change any of them; the options and quotation follow. Select a source to see it in the email.
- 720 W
- 40 W
- 300 W
- 150 W
- 600 W
- 2,000 W
Runtime
Budget incl. VAT
Matching options: 1,210 W load, 4 hours
Quotation Q-0142 · draft
Backup power for Van Wyk & Mokoena Attorneys
Pretoria East · valid 14 days · prepared by Highveld Power Supplies
Total incl. VATR82,570
| Item | Qty | Unit | Amount |
|---|---|---|---|
| 3 kVA hybrid inverter INV-3K | 1 | R14,950 | R14,950 |
| 10.2 kWh lithium battery BAT-10 | 1 | R44,500 | R44,500 |
| Installation kit (DB board, isolators, cabling) KIT-STD | 1 | R3,850 | R3,850 |
| Installation and commissioning, Gauteng SVC-INST | 1 | R8,500 | R8,500 |
| Subtotal | R71,800 | ||
| VAT 15% | R10,770 | ||
| Total | R82,570 | ||
To confirm with the customer before sending
- Is the office supply single-phase or three-phase?
- Where is the distribution board, and is there wall space beside it for the inverter?
- Do any computers use a UPS today? (It can be retired once the system is in.)
7 catalogue items checked
Where it connects
- Knowledge
- Product specifications, datasheets, compatibility and the sizing rules your specialists use.
- Rules
- Pricing, discounts, bundles and when a quote needs a manager's approval.
- Systems
- Your catalogue or ERP for stock and price, your CRM for the customer, your quoting tool or email to send.
How it's built
Reading the enquiryLanguage models
A language model turns the customer's free text into structured requirements, each tied to the words it came from, so the salesperson can see exactly why the application read it that way.
Finding the right productsRetrieval (RAG) and rules
Retrieval-augmented generation (RAG) searches your specifications and datasheets to narrow the catalogue. Your sizing and compatibility rules then decide what qualifies, so the application never offers a product or price that isn't in your catalogue.
What belongs togetherGraph analytics
A graph of products, accessories and compatibility adds what each option needs, such as the right installation kit, and flags combinations that don't work together.
Running the stepsLangGraph and LangChain
LangGraph runs the work as a defined sequence (read, size, match, check stock, draft, ask), with a review step before anything is sent. LangChain connects the model, the retrieval and your systems.
Where it runsYour environment
In your cloud account, or on your own infrastructure where your data can't leave it.
For operations teams handling applications and supporting documents
Get customer applications ready for a decision
Less chasing, re-checking and re-typing between the first email and an application your team can decide on.
What arrives
Messages, forms and attachments: application forms, certificates, letters and statements, often incomplete.
What the application does
Captures every detail with its source, checks it against the other documents and your policy, and flags what's missing, out of date or inconsistent.
What your team receives
A complete application record, the outstanding items, a follow-up ready to send, and a summary for the decision-maker.
What happens next
As the customer replies, the record updates. When nothing is outstanding, it moves to review in your existing system.
Trade account applications
Sample workspace · Karoo Fresh Foods
- Received
- Captured
- Waiting on customer
- Ready for credit review
What arrived
From Lindiwe Khumalo, Mzansi Corner Market
Email: Opening a 30-day account
Received 2026-10-03
- Good day,
- We'd like to open a 30-day account so we can stock your dairy and fresh range at our two stores.
- Our credit application and company documents are attached.
- We're asking for a R50,000 limit.
- VAT no. 4123456789
- Lindiwe Khumalo, Owner, Mzansi Corner Market
Application record
7 of 10 completeSelect a field to open the document it came from.
Outstanding before review
VAT number doesn't match
The application form and email give 4123456789. The VAT registration notice gives 4123456798.
Bank letter is out of date
The bank confirmation is dated 12 May 2026. The credit policy needs one from the last 3 months.
Director's suretyship missing
Limits above R25,000 need a signed suretyship from a director.
Required documents
- Signed credit application
- Company registration certificate
- VAT registration notice
- Bank confirmation, last 3 monthsout of date
- Director's suretyship (limit over R25,000)missing
Draft email to Lindiwe Khumalo · for review before sending
Hi Lindiwe,
Thank you for your application. To finish setting up your account, we need:
- Please confirm your VAT number: your form shows 4123456789, but your VAT notice shows 4123456798.
- Please send a bank confirmation letter dated within the last 3 months.
- Please sign and return the attached director's suretyship.
Once we have these, your application goes straight to our credit team.
Where it connects
- Knowledge
- The document types you accept, and what each one must show.
- Rules
- Your credit or onboarding policy: limits, validity periods, required security and approval levels.
- Systems
- The inbox or portal where applications arrive, your document store, and your CRM or ERP for the new account.
How it's built
Reading the documentsLanguage models
A language model reads each message and attachment into the fields your team needs, and quotes the exact text each value came from. A value without a matching quote isn't accepted.
Checking against your policyRules
Your policy becomes explicit checks: required documents, how recent each must be, and values that must agree across documents. Every flag says which rule it comes from.
Recognising who's whoGraph analytics
A graph links applicants, companies, directors and documents, so the same company or person is recognised across applications and accounts.
Holding the application openLangGraph and LangChain
LangGraph keeps each application as a live case: it pauses for the customer's reply and for your team's review, then picks up where it left off. LangChain connects the model to your inbox, documents and systems.
Personal informationPrivacy
Only the details a step needs are passed to the model, and processing can stay inside your environment in line with POPIA.
Have a task your team repeats every day?
Tell us what arrives, what your team does with it, and what goes out. We'll show you what an application for it would look like.