SAP to open-stack ERP migration
Manufacturer, 500+ employees. Phased migration off SAP ERP onto an open stack with custom modules — run in stages while the plant kept working.
- 0 downtime
- 8 mo migration
- −40% on licences
Generative AI features that survive the compliance review
Drafting, extraction, summarisation and classification — with the boring parts included: human review where it matters, evaluation you can rerun, and a cost ceiling your finance team signed off.
What we can show you
Overview
Generative AI, past the demo
Generative AI reads and drafts well, and decides badly. Every feature we ship keeps a human where the stakes are real.
What generative AI actually does in a business process
Generative AI produces text, structure or classification from unstructured input: an invoice becomes rows, a contract becomes a summary, a support thread becomes a routed ticket. It is good at reading and drafting, and unreliable at deciding — which is why every feature we ship has a human in the loop where the stakes are real.
The engineering is not the model call. It is the grounding, the review queue, the evaluation set and the cost ceiling that turn a demo into something a compliance officer will sign off.
When a generative feature is justified
When the same unstructured input arrives in volume and a person currently reads all of it. If the volume is ten documents a week, a person is cheaper than a project — and we will say so before you spend anything.
Signals a process is ready for generative AI
- someone reads the same kind of document dozens of times a day;
- the output is structured — fields, categories, a summary with a fixed shape;
- accuracy can be checked by a reviewer in seconds rather than minutes;
- the source documents already exist in digital form;
- a wrong answer costs a correction, not a lawsuit.
Price
From $10,900
Timeline
3–5 weeks
Payment
50/50 by milestone
Comparison
Generative feature or an off-the-shelf tool
Where a subscription tool is enough, and where it stops being enough.
Criterion
Ready-made toolGenerative feature we build
Built for youOff-the-shelf AI tool
Grounded in your data
Whatever you paste in
Your systems, with permissions preserved
Review workflow
Copy out, paste back
Queue with confidence thresholds, built in
Audit trail
None you can export
Prompts, outputs and approvals stored
Cost control
Per seat, per month
Per request, capped in code
Data residency
Vendor decides
Local or region-pinned models when required
When it wins
Ad-hoc drafting by individuals
A process with volume, rules and an owner
Scope
What we build
Four generative patterns that pay for themselves, and one interface that decides whether they do.
Document extraction
Invoices, contracts and forms into structured data, with confidence scores and a review queue for anything below the threshold.
Drafting
Proposals, replies and summaries seeded from your own past work. Nothing is sent unreviewed — the draft is the product.
Classification and routing
Intake triage that explains its reasoning, so the person who disagrees with it can see why it decided that.
Content operations
Product descriptions, translations and metadata at catalogue scale, with sampling-based quality control instead of blind trust.
Human-in-the-loop is the design, not the fallback
Confidence thresholds decide what auto-processes and what queues for a person. The review interface is part of the deliverable, because that is usually where the return actually lives: a reviewer approving in three seconds beats a model being right 92% of the time and accountable to nobody.
Feature in your product
From $10,900
Timeline
3–5 weeks
Full generative product
From $35,900
Governance
What your security reviewer will ask
We would rather answer these on a web page than in a questionnaire three weeks into procurement.
Controls we ship by default
Zero-retention API configurations · no client data used for training · PII redaction before the model call where the task allows it · a full audit trail of prompts, outputs and approvals · model choice that respects data residency, including local models · a monthly cost cap enforced in code.
Everything above is written into the data processing agreement, not just into a sales deck.
Why us
Why teams choose us for this
Four things we can prove, not four adjectives.
Human review is the design
Confidence thresholds decide what auto-processes and what queues for a person. The review interface ships with the feature, because that is where the return actually lives.
Evaluated before launch
A labelled set from your own documents, scored in CI. You see the accuracy number and the threshold behind it before anything goes live.
Governance built in
Zero-retention configurations, PII redaction where the task allows, and a full audit trail of prompts, outputs and approvals.
Cost capped in code
Inference is passed through at cost against a monthly ceiling you set — no token mark-up and no surprise invoice.
Industries
What we buildfor your industry
Document-heavy work is where this pays: logistics paperwork, insurance claims, property contracts and e-commerce catalogues.
- IT companies and digital agencies
- Insurers and brokers
- Lenders and microfinance
- E-commerce
- Wholesale and distribution
- Manufacturing
- Freight and forwarding
- Clinics and medical centres
- Dental
- Car dealers and service centres
- EdTech
- Legal and consulting firms
- Equipment rental
- Real estate agencies
- Property developers
- Travel and events
- Beauty and wellness
- Construction and renovation
FAQ
Generative AI questions
What corporate buyers ask before the pilot.
How is this different from letting the team use ChatGPT?
Grounding in your data, permissions, an audit trail, evaluation and cost control. A chat window is a tool; this is a feature with an owner and a test suite.
How accurate is extraction?
Task-dependent and measured on your documents. We publish the number and the confidence threshold that drives auto-processing, so you decide the trade-off rather than inheriting ours.
Can we keep the data in-house?
Yes, with local or region-pinned models. We will tell you the quality trade-off honestly rather than pretending there is none.
Who reviews the output?
Your people, in an interface we build, with the queue ordered by confidence so attention goes where it is needed.
What does it cost to run?
Modelled in week one and capped in code. Most features land in the tens to low hundreds of dollars a month; inference is passed through at cost.
What if the model provider changes their API?
The integration is abstracted: swapping providers is a configuration change plus a rerun of the evaluation set.
2–4×
faster on the code-writing share
in 3 weeks
from idea to MVP
230+
projects delivered
17 years
of engineering experience
Work
Work we can show you
SMARTWILL
Digital inheritance service (Web3)
AIRDEVELOPPA
Indoor air quality finder driven by sensor data
CAR5
Car-sharing app
APP.DENTAL
Cloud CRM for dental clinics
NEXORA
Warehouse and logistics management platform
As-built docs
Digital as-built documentation
EX211
Telegram Mini App for currency exchange
SMARTWILL
Digital inheritance service (Web3)
Start with one document type
Pick the highest-volume, most annoying document in your process. One week of feasibility tells you the accuracy, the cost per document and whether it is worth automating at all.
If the honest answer is "a template and a macro", you will hear that.