Microsoft 365 AI automation
Copilot Studio and Power Automate AI automation, built in 5 to 10 days
We build Copilot Studio agents and Power Automate flows on the SharePoint, Outlook, Teams, and Excel you already use. Fixed price, fixed scope. Every build ships with written acceptance tests and a 14-day warranty.
Built by Narek Maloyan and Tigran, a founder-led team in Yerevan, Armenia. Narek built the agent that reached #1 on SWE-bench Verified (70%, May 2025), and has 7+ years of production machine learning and a PhD. You talk to the engineers who build your project.
Two packages, one workflow each
Half at kickoff. Half when the acceptance tests pass.
Package A
SharePoint knowledge agent
$3,000 to $4,000 · 10 working days
A Copilot Studio agent that answers staff questions from your SharePoint documents. It cites the source file and respects who can see what.
What you get
- An agent built in Copilot Studio on your SharePoint sites and libraries
- Answers with a link to the source document
- Sign-in with Microsoft, so each user only gets content they can already open
- Published to Microsoft Teams
- A test set of 20 to 30 real questions, written with your team
- A handover document and a 1-hour session for your admin
Acceptance tests
- At least 90% of the test questions answered correctly, each with a source link
- Questions the documents do not cover get "I don't know", never an invented answer
- A user without access to a file never gets content from it, checked with 2 real users
- Your admin can add or remove a knowledge source without us
Package B
Document to Excel flow
$2,000 · 5 working days
A Power Automate flow with AI extraction. Invoice or contract PDFs land in SharePoint. Checked data lands in Excel. Anything doubtful goes to a person first.
What you get
- Trigger on a SharePoint library or an Outlook mailbox
- AI extraction of the fields you name (AI Builder or your Azure OpenAI)
- Business checks: totals add up, VAT rate and VAT ID valid, IBAN checksum, dates, duplicates
- A confidence score per field and a review queue for low scores
- Approval in Teams and Outlook, then the row is updated in Excel Online
- Error alerts to your admin, with a link to the failed run
Acceptance tests
- Run on 30 of your real documents: every field we agree as required is checked
- Every planted error in a test batch (wrong total, bad VAT ID, duplicate) reaches the review queue
- No document with a failed check is written as approved
- A correction in the review queue is checked again before it is accepted
14-day warranty. For 14 days after acceptance, we fix for free any case where the build fails an acceptance test on the same kind of input it passed on. No hourly billing, no change request.
From call to accepted in two weeks
- Day 0, scoping call. 30 minutes, free. We agree on one workflow, the fields or questions, and the acceptance tests in writing.
- Day 1, access and test set. You give us a test account and sample documents. We work in a test environment first, never on live data.
- Build, daily updates. A short written update every day, with what works and what is next. You can try it from day 3.
- Last day, acceptance. We run the acceptance tests together. You pay the second half only when they pass.
Working demo: 6 of 6 planted errors caught
We ran the Package B pipeline on a test batch of 13 synthetic invoices in 4 layouts, 6 of them with a planted error. An LLM and a rule set read each PDF, and 11 business checks run on the result.
The invoices are synthetic. Your own documents get their own accuracy report before you pay the second half. Read the full write-up: AI invoice checker demo.
Who builds it
- 50+ Power Automate flows built in 2026 for a financial services company, plus Power Pages apps, Copilot Studio agents over PDFs, and Outlook and SharePoint flows. All generated from code, so every change is versioned and reviewed.
- #1 on SWE-bench Verified (70%, May 2025). Narek Maloyan built that agent as Founding AI Engineer at Zencoder.
- 7+ years of production machine learning at Viasat and TrendMD, and a PhD in Computer Science.
You work with the engineers who build it, not an account manager. Client names and data from past work stay private.
Questions
What do we need on our side?
A Microsoft 365 tenant, a test account for us, and sample documents or questions. The knowledge agent needs Copilot Studio or Microsoft 365 Copilot licenses. The document flow needs Power Automate Premium or AI Builder credits. We tell you exactly which on the scoping call.
Where does our data go?
It stays in your tenant. The agent and the flow run in your Microsoft 365 and Power Platform environments. If you want an external model, it is your own Azure OpenAI resource.
Do you work with Microsoft partners?
Yes. Partners can resell either package and we deliver under your brand, with your client in your tenant.
What if the workflow is bigger?
We start with one workflow at a fixed price. When it passes its tests, we quote the next one.
Related: RAG knowledge bots outside Microsoft 365, invoice chasing, and how we handle your data.
Pick one workflow. We agree the tests on a 30-minute call.
Book a scoping call. You leave with the workflow, the acceptance tests, and a fixed price in writing.