Pack 01
live
live
Higher education
Admissions, counselling and student operations. Live across more than eleven institutions — the sector that taught us what belongs in the engine and what belongs in a pack.
Enterprise AI Engineering
Engineering
Engineering
Intelligent Enterprises.
Businesses will not win because they use more AI. They will win because they become more intelligent — and that is an engineering problem, not a procurement one.
11+Institutions running our first sector in production
DaysTo a configured agent, once a pack exists
WeeksTo an entirely new industry
The market as it stands
Everyone is selling construction.
There are four ways to buy enterprise AI today, and three of them end in a bespoke build. The agency constructs your system from scratch. The consultancy recommends that someone construct it. Hiring in-house means constructing it yourself, after twelve months of recruiting.
All three price the same work over and over. The second institution in a sector pays nearly what the first paid, because nothing was built to be reused. The industry competes on who constructs faster and cheaper — which is a race, not a strategy.
We think construction is the wrong unit of work.
The thesis
One engine. Many packs.
A sector-neutral core that has never heard of a student, a patient or a supplier — and thin packs that carry everything a sector actually means.
This is the whole strategy. Because the domain lives in data rather than in code, the second institution in a sector is configuration. Because the engine is shared, every sector inherits the grounding, the guardrails and the eval suite that the previous sector paid for.
The usual trade-off says you may have depth or you may have economics, and never both — bespoke work is deep and expensive, platforms are cheap and shallow. The engine/pack split refuses that trade. Depth lives in the pack. Economics live in the engine.
The compounding
- Sector one pays for the engine
- Sector two pays for a pack
- Every fix to the engine reaches every sector at once
- Every eval written stays written, permanently
Where we sit
A different shape, not a better score.
Plotted against what this industry competes on. The point is not that our line is higher — it is that it moves in the opposite direction.
Doing AI
AI agency
Strategy consultancy
Our reading of the alternatives, not an audit of any named firm. We score deliberately low where the industry scores high — bespoke construction, advisory, and the cost of the second sector. Those are the factors we are trying to remove from the category, not win on.
What we removed, what we raised
The trade we actually made.
Eliminated
- Bespoke construction per client. Replaced by pack configuration on a shared engine.
- The strategy phase. No discovery deck, no roadmap deliverable, no readiness assessment.
- Pilots and proofs of concept. If we cannot see production, we decline the work.
- The account layer. No delivery team you meet after signing.
Reduced
- Breadth of AI menu. We do not offer everything AI can do. We do operational systems.
- Custom model work. Almost always the wrong lever; the value is in grounding and orchestration.
- Configurability at the edges. Packs are opinionated on purpose.
Raised
- Speed to production. Days for a configured agent once a pack exists; weeks for a new sector.
- Grounding. Answers come from your source of truth with citations, or the system escalates.
- Ownership. Your cloud, your data, exportable, documented.
- Accountability after launch. We operate what we shipped.
Created
- The engine/pack split as a purchasable thing. Sector neutrality enforced in CI, not promised in a meeting.
- Eval suites built from real conversations — shipped as an artifact you keep, and the gate every release must pass.
- A sector roadmap. Buy pack one knowing pack two is weeks away, not a fresh negotiation.
The roadmap
Sectors, in the order they ship.
Each one is a pack on the same engine. The distance between them is weeks, and it shortens every time.
Pack 02
in build
in build
Manufacturing
Quality control, exceptions and shop-floor status. Grown out of a QR-code inspection system built for a vehicle assembly line.
Pack 03
next
next
Healthcare & clinics
Intake, scheduling, clinical documentation. Same ontology shape as admissions; different words, different regulator.
Pack 04
next
next
Financial & professional services
Document review, proposals, client assistance, with the audit trail the sector requires.
Pack 05
mapped
mapped
Commerce
Retention, catalogue operations, customer contact.
Pack 06
mapped
mapped
Logistics
Forecasting, purchase orders, shipping documentation, live status.
Invariants
Four rules the code enforces.
Not values on a wall. Each of these is a build gate, and breaking one fails the release.
Grounded, or silent.
An agent answers from the tenant's source of truth, with a citation, or it escalates to a person. It does not improvise. In higher education that rule has a name — no invented fee ever reaches a family.
Evals gate every release.
Fast deployment is only honest when a suite catches the regression first. No evals, no ship. The suite is built from real conversations and it stays yours.
The engine never learns your vocabulary.
Domain words live in packs. grep the engine for student and every hit is a build failure. This is what keeps a new industry a configuration rather than a rewrite.
Official channels, India-region, by design.
Sanctioned platform APIs only — an unofficial integration gets an institution banned within weeks. Data resident in India, DPDP-aligned from the schema up rather than retrofitted.
Who this is for
Mostly people who already said no.
You ran a pilot and it stalled.
Something worked in a demo and never reached production. The gap was never the model — it was grounding, permissions, evals and somebody owning it on a Sunday.
You priced a bespoke build and walked away.
Too slow, too expensive, and at the end of it you would not have owned the thing. A pack costs a fraction of a first build because the engine is already paid for.
You assumed this was priced for the Fortune 500.
Most of our work is with mid-sized institutions — universities, clinics, manufacturers — who were never the customer enterprise AI was designed and priced for. That is the market we are actually building for.
Evidence
What exists today.
Higher education, in production
An operating system for university administration, live across more than eleven institutions. Pack one, and the sector that taught us where the seam between engine and pack belongs.
Manufacturing, on the floor
A QR-code quality-control system built for a vehicle assembly line — the origin of pack two.
Two commercial products
LEOS, an AI sales agent, and BillGill, a sales-intelligence system. Both built end to end.
The engine itself
Sector-neutral core with the packs, evals and tenancy model described above. Built on LangGraph, Postgres with pgvector, LiteLLM and Claude.
We do not publish client results we cannot evidence, and we do not present one engagement's range as a benchmark. On a call we put measured numbers against your own workflow — including where this will not help.
Which sector are you in?
If it is one we have a pack for, you could be in production this month. If it is not, you would be the sector that defines the next pack — and that is a different, better conversation.