Case studies
Systems we have built, described by what they do.
Real builds, described by solution type. No client names and no invented results: measured outcomes are added only when they are measured.
Case study · Operations System
One record from job request to invoice.
A field-service operations platform for HVAC, electrical and plumbing teams.
The problem
Field teams ran jobs across phone calls, WhatsApp and spreadsheets. Jobs were lost between the office and the field, completion proof was scattered across personal chats, and invoicing waited until someone had time.
What we built
One operations platform covering job intake, scheduling and dispatch, mobile job completion with photos and signatures, quotes, invoicing and finance reporting.
- 01Request
Every call, message or form becomes a job with a customer, address and owner.
- 02Dispatch
Dispatchers see the day on one board and assign the right technician.
- 03Complete on mobile
Technicians record notes, photos and the customer signature on their phone.
- 04Approve
The office reviews completed jobs and sends back anything incomplete.
- 05Invoice and report
Approved jobs move straight to invoicing; finance sees billed, paid and overdue.
Next.jsSupabasePostgreSQLRole-based accessPlaywright tests
Operations System: from US$14,000Case study · AI Knowledge Assistant
Answers you can check, from documents that change.
A retrieval-augmented knowledge assistant for a financial-services product range.
The problem
Customers and staff asked detailed questions about card products: fees, eligibility, benefits and terms. The answers sat in long documents that changed often, and a wrong answer was costly.
What we built
An assistant that answers only from the approved product documents, shows the source for every answer and declines when the documents do not cover the question.
- 01Prepare
Approved documents are split into passages and indexed with their source.
- 02Retrieve
The most relevant passages are found for each question.
- 03Answer
The model writes the answer from those passages only, and cites them.
- 04Decline
If the passages do not answer it, the assistant says so instead of guessing.
- 05Evaluate
Real questions are run before every release to check accuracy.
PythonLangChainVector storeLLMEvaluation harness
AI Knowledge Assistant: US$6,500 + US$550 a month
How it works
From first message to a working system.
- 01Send a briefDay one
- 02Review and recommendationWithin a few days
- 03Fixed scope and proposalBefore any build
- 04Build in weekly releasesTypically 2–8 weeks
- 05Launch and handoverLaunch week
- 06Support and improveOptional, monthly
Start a project
Tell us what needs to work better.
Share the workflow, product or operational bottleneck. We’ll review the context and recommend the most practical next step.
Email usinfo@sequenceautomate.comAsync-first. Available worldwide.
No meeting required to send your brief.