Six engagements, including our own.

Cintram — client delivery platform

20+

Product modules

3

Client surfaces

15 – 22

Angular

4.2 – 5.2

Django

The problem

Service businesses — consultancies, coaching and training practices — sell an engagement, then deliver it over weeks or months across a spreadsheet, a shared drive and a long email thread. Nobody can answer “where is my project” without a person going to look, and the client’s experience of a professional firm ends the moment the contract is signed.

What we built

A platform where a closed deal becomes a project with its phases and tasks already in place, and where the client sees the same truth the team does — through a white-labelled portal carrying their brand, not ours. Around twenty product modules sit behind one API contract: CRM and contacts, leads, opportunities and sales pipeline, projects, planner and tasks, an LMS for training delivery, marketing and forms, members and teams, visitor tracking, automation, and recurring billing with mid-term proration. The API is described in OpenAPI and the Angular client’s types are generated from that schema, so a breaking backend change fails the frontend build rather than a customer’s screen.

Where it is now

In production and actively developed. The frontend has been carried from Angular 15 to Angular 22 and the backend from Django 4.2 to 5.2 — incrementally, in a live product, without a rewrite or a migration weekend. That is the same modernisation work we do for clients, run on our own codebase where the consequences land on us first.

cintram.com — project delivery

Credit underwriting copilot

-71%

Analyst review time

$4.2M

Annual saving

11 wks

Brief to production

The problem

Analysts were spending around six hours per deal reading filings before they could write the first line of a credit memo. The queue was long enough that the firm was losing transactions to faster competitors, and hiring more analysts had already been tried.

What we built

A retrieval layer over twelve years of filings and internal credit policy, an agent that drafts the memo section by section with a citation back to the source page for every assertion, and a review interface where an analyst accepts, edits or rejects each paragraph. Nothing reaches a committee without a human signature.

The result

Drafting fell to under ninety seconds, with review time down 71% overall. Analysts retained final authority on every claim — the condition the risk committee set before approving the project at all.

Remote patient monitoring

1.2M

Patients onboarded

-46%

Avoidable readmissions

40

Clinics live

The problem

Forty clinics were tracking post-discharge patients through phone calls and spreadsheets. Deterioration was routinely noticed days late, and no clinician could see which of the day’s patients most needed attention.

What we built

A mobile application pairing with home vitals devices, a triage model running on the device so it still works without a signal, and a clinician dashboard that orders the day’s caseload by clinical risk. The full data path is HIPAA-audited, with consent and access logged end to end.

The result

Nurses now see the highest-risk patients first. Avoidable readmissions across the participating clinics fell 46% over the first eighteen months.

Fleet routing platform

-38%

Fuel cost

-61%

Compute spend

30s

Replanning cycle

The problem

Routes for twelve thousand vehicles were planned overnight and then invalidated by nine in the morning. The existing solver took too long to re-run, so dispatchers overrode it manually and the optimisation was effectively decorative.

What we built

A solver rewritten to re-plan the whole fleet every thirty seconds against live traffic and fuel prices, checkpointed so it can run on interruptible spot capacity — a reclaimed node costs seconds rather than an entire run. Dispatchers keep manual override, but now rarely use it.

The result

Fuel cost down 38% and compute spend down 61% against the previous overnight batch, with dispatcher overrides falling to under 4% of routes.

Commerce & demand forecasting

+27%

Conversion

-33%

Dead stock

0.9s

LCP on 4G

The problem

A slow storefront was losing mobile customers, and buying decisions were being made in a spreadsheet three steps removed from the sales data. Overstock was written off every season as a cost of doing business.

What we built

A rebuilt headless storefront with a performance budget enforced in CI, and a demand-forecasting model surfaced directly in the purchasing screen — buyers see projected demand and current cover in the same view where they commit an order.

The result

Conversion up 27%, end-of-season write-offs down a third, and a storefront that holds under a second to largest contentful paint on a 4G connection.

High-stakes assessment platform

2.4M

Examinations delivered

99.99%

Exam-window uptime

300

Institutions

The problem

Examination load arrives in sharp, immovable spikes, and an outage inside the window is not a bug — it is an incident with a regulator attached. The incumbent platform had failed twice in two years.

What we built

A platform capacity-tested against three times peak load, with adaptive question selection and an assisted marker for long-form answers that always shows the rubric line it scored against, so a moderator can audit any grade in seconds.

The result

2.4 million examinations delivered with 99.99% uptime across exam windows, and marking turnaround reduced from weeks to days without removing moderator oversight.