Skip to content
Sortup

Production AI engineering

Most AI projects die between the demo and production. We build the part that ships.

Agents, RAG systems and the web and mobile apps around them, with evals, monitoring and code you own.

passed

Illustrative
  1. 0 ms

    Question

    What is the refund window for reseller annual plans?

  2. 180 ms

    Retrieval

    3 sources · permission filter: sales-ops

    • ▸ Refund policy v4.pdf · §3.2
    • ▸ Reseller agreement (EU) · §7
    • ▸ Notion · Billing FAQ
  3. 410 ms

    Tool call

    contracts.lookup(region="EU", channel="reseller")

    • → refund_days: 30, then prorated
  4. 1.31 s

    Eval checkpassed

    Grounded in sources · 3 of 3 citations verified

  5. 1.62 s

    Answer

    30 days from purchase [1][2], then prorated by month [1]. The reseller processes it [2].

latency 1.62 s · 2,140 tokens✓ within budget

Projects shipped
40+
Engineers
15
Team shipping since
2018
Sortup since 2020
Lahore
UTC+5

The five layers

Every AI product has five layers. You rent one.

  • You rent
  • We build
  • Where demos stall

Why AI demos stall

Knowledge Assistant run, layer by layer · Illustrative
  1. 05

    The product

    1.62 sCited answer · Slack #sales-ops

    Demos stall here:No product around the model

  2. 04

    Evals and guardrails

    1.31 s✓ 3 of 3 citations verified

    Demos stall here:No evals

  3. 03

    Agent logic

    410 mscontracts.lookup(region="EU")

  4. 02

    Data and retrieval

    180 msRefund policy v4.pdf · §3.2 · permission: sales-ops

    Demos stall here:No data pipeline

  • 01

    Model

    OpenAI · Anthropic · Google · open weights

Start here

Two systems, live in 14 days. Then decide.

All services

Client work

Real clients. Live products you can open.

All client work

We publish results only once a client confirms them.

The team delivered a complex super app with e-wallet, e-commerce, logistics integrations, and a dynamic commission engine. Their communication was clear, execution was reliable, and they consistently provided solution-driven results throughout the project.
Simon SprouleCo-founder, Gather

How it runs

The output of an in-house AI team. Without the six-month hire.

  1. 01 · 20 minutes

    Fit call

  2. 02 · Days 1–2

    Scope

  3. 03 · 2-week milestones

    Build

  4. 04 · Final week

    Launch and handover

Engineer reply
Within 1 business day
Code and repository
Yours from commit 1
Staging link
Every 2 weeks
Eval set
Agreed on day 1
Written update
Every week
NDA
Yours or ours, before the call
After launch
We fix what we shipped, 30 days
Handover
Runbooks, docs and a call

Before you ask

Will you work our hours?
We work your hours: the engineers on your project keep your working day, whether you are in the GCC, the UK and EU or the US.
What happens to our data?
It stays in your accounts where possible, under API terms that exclude training. We sign your DPA.

All questionsCompare with hiring, freelancers and agencies

Book

Book a 20-minute fit call.

No pitch deck. An engineer, and a straight answer.

Loading the calendar…

Already decided?

Send your brief

About five minutes. It saves as you type.

Not ready for a call?

Get a free AI Teardown

A 2-page build plan and a clickable prototype, in five business days.

Under your NDA if you want one.

Prefer email? hello@sortup.dev