How we work

The people who scope the work are the people who build it.

The person who writes the proposal writes the code. Every week you get a written note with what was built, what was decided, and why.

Four phases: diagnose, prove, build, hand overA written note every FridayTwelve weeks, gate to gate

The shape of an engagement

Diagnose, prove, build, hand over.

Weeks 1–2
Diagnose

Paid discovery sprint. Read the code, meet the people, write the findings. You keep the document whatever you decide.
Weeks 3–6
Prove

Eval set and baseline first, then the thinnest end-to-end slice in production. Cost model published before we scale anything.
Weeks 7–10
Build

Fortnightly increments. Same people throughout. ADRs and demo notes in your repo, not ours.
Weeks 11–12
Hand over

Your team runs it for two weeks while we are still on call. Runbooks, on-call rota, and a written exit review.

01Two founders on every engagement: one writes the code, the other owns whether it moves your numbers.
02Fixed team for the engagement. Substitutions need your written consent.
03Everything we produce is in your repositories and your cloud accounts from day one.

04Weekly written demo note. If it cannot be written down, it is not done.
05Evals and cost models are deliverables, reviewed with the code.
06Exit review is published to you whether or not we are re-engaged.

A week with us

Monday note, Thursday demo, Friday decision.

The same week, every week of the build. Nothing waits for a steering committee.

Fig. 01A week with us. You see the plan on Monday, running software on Thursday and the written note on Friday; the system runs its own evals on every pull request in between.

The first ten days

From your first message to the first weekly note.

Every line here is something we control, which is why we can put it in writing.

d01 You write a paragraph. A founder replies within one working day.
d02 30-minute call with both founders. No deck.
d04 Written read of your problem: what we would build, roughly what it costs, whoever you build with.
d06 Discovery sprint starts, if you want it. Repo and wiki created in your accounts.
d10 First weekly note in your inbox: what was built, what was decided, and why.

What you receive

The weekly note, as it arrives.

The written things are the deliverables. This is the one you will read most often: one page, every Friday, from the person who did the work.

Weekly demo note Week 07 of 12

Claims triage · week 7: second depot of handlers on the new routing

Built

Confidence band now applied per claim type; motor-glass and windscreen route straight through above 0.82.
Handler view shows the model’s evidence in the handler’s own words; reassignment reason captured on every override.

Decided

Secondary fraud model stays in-house rather than third-party: latency and data residency. ADR-011.

Measured

Held-out set, 2,1xx claims: precision 0.xx, recall 0.xx, unchanged from week 6 threshold.
Cost per claim £0.xx against ceiling £0.xx; forecast for full volume within ceiling.

Next, and one question for you

Week 8: drift alert wired to the on-call rota; parallel run begins for region.
Do you want complaints-rate measured per handler team or per claim type at gate 4? We recommend per claim type.

1One page, every Friday, written by the person who did the work. If it cannot be written down, it is not done.
2Every decision carries its reason and a link to the ADR in your repository.
3Numbers from the eval run on this week’s pull requests, not from a demo set.
4Cost model re-forecast every fortnight against the ceiling in the statement of work.
5The question we need you to answer before Monday, so the gate is never a surprise.

Fig. 03Sample weekly demo note. Built, decided, measured, next, and the one question we need answered before Monday. Values redacted; the format is the point.

Engagement models

Three ways to work with us. Same people in each.

Fixed price for scoped work, billed by milestone. A fixed itemised quote follows the discovery call; there is no hourly billing.

Engagement models compared
	Discovery sprint	Fixed-scope build	Embedded founders
Best for	Deciding whether and what to build	A defined outcome with a defined end	Ongoing senior capacity alongside your team
Duration	2 weeks	8–14 weeks	Monthly, 30-day notice
Who	Both founders	Both founders, for the whole engagement	One or both founders, part-time or full-time
Commercials	Fixed fee, quoted after the first call	Fixed price, milestone-billed, quoted after discovery	Monthly retainer
You receive	Findings document, prototype, costed build plan	Production system, ADR log, evals, runbooks, exit review	Weekly demo notes, quarterly architecture review
Change control	n/a	Written change note, re-priced within 3 working days	Re-prioritised in the weekly planning note
IP	Yours from day one	Yours from day one	Yours from day one

We suggest starting with a discovery sprint. Its fee is credited against a fixed-scope build signed within 60 days. We invoice in INR, or in USD, GBP, EUR or AED at the RBI reference rate on request.

No hourly billing · GST zero-rated as export of services; 18% GST for India-billed clients

Commercials

Fixed price, milestone-billed. No large lump sum up front.

A fixed-scope build is billed as a share of the agreed fee at each gate. Each gate is a written decision you make with working software in front of you.

Milestone billing for a 12-week fixed-scope build
Milestone	Weeks	Share of fee	Billed when
Discovery sprint (credited if already done)	Weeks 1–2	10%	On signature
Eval set, baseline and cost model	3–4	20%	You accept the baseline
Thin slice live in production	5–6	25%	First real traffic
Build in fortnightly increments	7–10	30%	Each increment demoed
Hand-over and exit review	11–12	15%	Your team has run it for two weeks
Total	12	100%

Included

Both founders for the whole engagement
Remote by default. On site by arrangement.
Two weeks on call after hand-over, no charge.
NDA before you share anything sensitive

Not included

Cloud hosting and model inference, which run in your accounts
Scope changes, priced by written change note within 3 working days

Payment terms 30 days. Optional monthly care plan after exit, with written SLAs, covering monitoring, patches and small improvements. You can stop after any milestone and keep everything built to that point.

Twelve weeks

What the calendar looks like.

Gates at weeks 2, 4, 6, 10 and 12. Each one is a written decision you make with the evidence in front of you.

Build Verify Cut-over Sign-off

Fig. 02A 12-week fixed-scope build. Verification runs in parallel from week 6; your team runs the system alone for the last fortnight while we shadow on call.

Fit

When we are the right choice, and when we are not.

Each alternative is right for something. The most useful thing a small firm can tell you is when to hire someone else.

Right for

01A scoped AI, data or product problem where judgement matters more than headcount
02You want the people who wrote the proposal to write the code
03You need evals, a cost model and an exit review as deliverables, not slides
04Two senior people for a quarter, not forty for a year

Wrong for

01You need fifty engineers by next month. A global SI can staff that; we cannot and will say so.
02You want a vendor to own the system after launch. We build in your accounts and hand over; we can stay on retainer, but we do not hold the keys.
03You want a proof of concept with no path to production. We will suggest the cheapest honest path, which is sometimes "not yet".
04A configured off-the-shelf tool fits your process. We will tell you on the first call, and help you evaluate it.

At the end

The exit review, written whether or not we are re-engaged.

What we got wrong comes first. Where everything lives comes second. It is the document the next engagement starts from.

Exit review Week 12 of 12 · page 1 of 6

What worked, what did not, and where everything is

Summary

Scope delivered as agreed at gate 2; two metrics agreed at gate 2 were met, one was missed and is explained below. Your engineers have run the system unaided since week 10.

What we got wrong

We underestimated the re-labelling effort after the week-6 threshold change: xx handler-hours, not the xx we planned. Budgeted in the runbook from now on.
Parallel run should have started one week earlier; the region roll-out slipped four working days.

Where everything is

Eval harness and labelled set: repo/evals/ · Cost model: repo/cost/model.xlsx + telemetry dashboard · ADR-001 to ADR-014: repo/docs/adr/ · Runbooks and on-call rota: repo/docs/runbooks/

Hand-over status

On-call shadowed weeks 11–12: x pages, all resolved by your team without escalation. Credentials rotated; our access revoked on date.

1Delivered in the final week whether or not we are re-engaged. It is a term in the statement of work.
2What we got wrong is written first, with what it cost. The point of the document is the next engagement, yours or anyone’s.
3Where every artefact lives, by path, in your accounts. The hand-over checklist is the table of contents.
4Your team has already run it alone for two weeks by the time this is written.

Fig. 04Sample exit review, first page. Misses are recorded with their cost; every artefact is listed by path in your accounts.

The terms

In every statement of work.

Terms in every statement of work

01 100% Code ownership, handed over Full source, documentation, credentials and deployment access are yours at launch. Any team can take over.
02 1 day Maximum response time Every message answered within one working day, by a founder, not a ticket queue.
03 0 Substitutions without consent The people who scope the work build it. Nobody joins or leaves the engagement without your written agreement.

Prepared for: Every client, in every statement of work

Abhinav RathiCo-founder · Technology & engineering

Abhishek RathiCo-founder · Product & strategy

Start with a two-week discovery sprint.

Fixed fee, credited against a build signed within 60 days. You keep the findings whatever you decide.

Book a discovery call Services