The pilot cannot be promoted.
It worked on a curated folder. Production means the real files, with their permissions, and nobody signs that off.
AI transformation · from one workload to your own Sovereign Artificial Intelligence (AI)
AI transformation in five stages, each ending in something you own
BiltIQ's AI transformation path is a five-stage plan for moving an organisation from one tested workload to its own sovereign AI platform: scope, pilot, install, act with approvals, and extend. Each stage ties to a BiltIQ product and ends in something the buyer keeps, starting with a paid pilot on the buyer's own documents.
We'll walk you through it on sample data.
ISO/IEC 27001:2022 certifiedISO 9001:2015 certified4 patent applications filed, none grantedWinner, NHA PM-JAY AI Challenge 2026 (hackathon)NVIDIA Inception PartnerDPIIT-recognised (DIPP239966)
Five stages
The AI transformation roadmap: five stages, five things you keep
The path:ScopePilotInstallAct, with approvalsExtend
| Stage | What happens | Product | You keep at the end |
|---|---|---|---|
| 1. Scope one workload | A scoping call: the owner, the documents, the rule and today's manual result. We walk you through the product on sample data. | Agent OS (walk-through) | A written next step and a baseline to measure against |
| 2. Pilot on your files | NDA first. Your AI runs for 4–6 weeks on past files with known outcomes, on your GPU server or a BiltIQ loaner. ₹5–8 L, ex-GST, credited against the licence. | Agent OS on AI Factory | A findings report against your manual results; the licence decision |
| 3. Install your platform | AI Factory and Agent OS go on your GPU server; hardware at cost. Site licence ₹50–90 L one-time, ex-GST. | AI Factory + Agent OS | Your own AI platform, with the managed service (₹16–18 L/yr, ex-GST) |
| 4. Act, with approvals | Your AI agents check, cite the page and propose. Irreversible and high-consequence actions pause for human approval. | Agent OS | Workflows with an approval record |
| 5. Extend | More workflows on the same platform, and Business OS for your books. Learning OS is offered as a hosted service from Apr 2027 (targeted), not on your platform; owning it on-prem is a roadmap item. | Business OS · Learning OS | More functions on your AI platform |
Why pilots stall
Four reasons AI pilots never reach production, and none is the model's fault
Answers nobody can check.
A fluent answer with no source is worse than none, because someone will act on it. Your AI cites the page.
Intelligence you rent.
With a rented cloud model, your usage runs on someone else's system. Stop paying, and the context leaves with the contract.
Ungoverned tools.
In many firms, staff already paste internal files into consumer tools. The choice was never between AI and none; it was governed or ungoverned.
Why decide now
Why the architecture decision is due now
DPDP Act 2023
Sets duties and penalties for how organisations handle personal data.
RBI FREE-AI
Recommends a board-approved AI policy.
SEBI AI guidance
The regulated entity answers for its models' outputs, even through third-party tools.
Public Procurement No. 1
Defence and government buyers check each model's origin.
An organisation choosing its AI platform now chooses the architecture it will be audited on later.
When not to buy
When owning your AI is the wrong call
If your volume is low, a metered cloud API can cost less than a GPU server, a licence and a yearly service. We work that out with you on the scoping call, on your numbers, and we say so when the answer is no. Our guide "on-prem vs cloud AI cost" shows the method.
Checkable, not claimed
Checkable facts behind our AI
- Winner, NHA PM-JAY AI Challenge 2026 (hackathon): claim and bill verification, IISc Bangalore finale, 8 May 2026.
- AI Factory installed on an engineering college's own GPUs (4 DGX Spark); setup by BiltIQ, unpaid.
- ISO/IEC 27001:2022 certified (certificate 305026060958IS, valid to 8 June 2029) and ISO 9001:2015 certified.
- 4 patent applications filed, none granted; the platform gate is application 202631094208.
- NVIDIA Inception Partner. DPIIT-recognised startup (DIPP239966).
- 12 people, 10 engineers, 80+ combined years. Paying clients in five countries.
- Our MIT-licensed privacy library detects Indian personal data such as Aadhaar and PAN and pseudonymises it.
The platform
One AI platform, four layers
GPU:
Your own GPU server, or our cluster in India (subscription: not yet on sale).
AI Factory:
Model serving. Single node per model today; multi-node written, not yet in use.
Agent OS:
AI agents, the platform gate and approvals. Outbound calls are blocked by default (platform gate). Irreversible and high-consequence actions pause for human approval.
Apps:
Agent OS workflows, Business OS, Learning OS.
Model-agnostic: open-weight models chosen per buyer. NVIDIA Nemotron by default for defence and PSU; Qwen3.8-27B for reasoning where allowed.
Four products
Four AI products, one platform
Agent OS
Documents, tenders, contracts, claimsReads, checks against your rules, cites the page, waits for approval
AI Factory
Models and GPUsServes open-weight models on your GPU server
Business OS
Accounts, GST, sales, stock, HRProposes changes; the 23 data-changing tools wait for a person
Learning OS
Courses and tutoringAI Teacher answers from the course book and cites chapter and page
Reading room
Reading room
What AI transformation actually means for an Indian enterprise
The five-stage AI transformation roadmap
How to staff an AI transformation
AI transformation in regulated sectors
Shadow AI: your staff are already pasting company data into chatbots
On-prem vs cloud AI cost comparison
FAQ
Frequently asked questions
What is AI transformation, in practical terms?
For BiltIQ, AI transformation means moving from one tested workload to an AI platform your organisation owns and can evidence. It starts with a scoping call and a paid pilot on your own documents, then installs AI Factory and Agent OS on your GPU server, puts AI agents to work with approvals, and extends to more functions on the same platform.
How long does an AI transformation take with BiltIQ?
The first measured step takes 4–6 weeks: a paid pilot on your past files, ending in a findings report and a licence decision. After that, timing depends on the workloads you choose and the state of your documents. BiltIQ quotes each later stage in writing, after the pilot, rather than promising a deployment date in advance.
When is on-premise AI the wrong choice?
When your volume is too low to repay a GPU server, a site licence and a yearly managed service, a metered cloud API can cost less. BiltIQ works through this on the scoping call using your own numbers and says so when owning your AI does not pay. The subscription tier, not yet on sale, is meant for that case.
Can we still use GPT or Claude alongside BiltIQ's AI platform?
That is your policy decision. In BiltIQ Agent OS, outbound calls are blocked by default by the platform gate, and an unconfigured install blocks all outbound calls. Outside on-prem-required mode, an administrator can allow specific outbound hosts. In on-prem-required mode, no outbound host can be allowed at all, so every model call stays on your servers.
AI agents with human approvalon-premise LLM platformAI ERP and CRM with GSTAI tutor that cites the bookon-premise AI pricingDPDP Act guide for AI
Start with the first stage: one workload, one call
We'll walk you through it on sample data.