Scale · AI and automation
Custom AI and GenAI, built to reach production.
Custom AI and generative AI (GenAI) for Indian enterprises whose pilots worked in a demo and stalled before production. NetScale tests one use case on the organisation’s own data, then builds it with controls on what the model may see, say and do. It deploys, monitors and maintains it, and hands the models and code over as theirs.
- Last updated
What it is
Custom AI and GenAI
Custom AI is software built around a model for one job in your business: a chatbot that answers customers, an assistant for staff, a model that reads images or text.
Generative AI, or GenAI, is the kind that writes answers, summaries and drafts. Most of the work is around the model: connecting it to your applications through APIs and microservices, deciding what it may see and say, and monitoring it once live.
AI governance and compliance controls are designed in from the start, not added after the first wrong answer.
Custom software that is not AI sits under Custom enterprise solution.
Staff use of third-party AI tools sits under Identity and access management (IAM).
Forecasts and predictions from your business data sit under Data and analytics.
What it covers
- Chatbots & custom models
- AI API integration
- Computer vision & NLP
- Deployment & lifecycle management
In this line: AI and automation
- Intelligent automation (RPA + AI)
- Intelligent document processing (IDP)
- Data and analytics
- Custom AI and GenAI solutions
- DocInflow document audit
Sound familiar?
- 01
AI pilots were run to prove it works here. None reached production. What stopped them, and is it still there?
Our answer: Custom AI and GenAI solutions
- 02
The assistant answered well in the demo. Who checks what it tells customers?
Our answer: AI governance and compliance controls
- 03
The model went live last year. Who has checked its answers since?
Our answer: Deployment, monitoring and lifecycle management
- 04
The AI vendor’s API is wired into one application. What happens when the vendor changes the model?
Our answer: AI API integration and microservices
What we do.
Chatbots, custom models, AI API integration, computer vision and NLP, with deployment, monitoring and lifecycle management.
- 01
Chatbots and knowledge-base assistants
Chatbots for customers and assistants for staff, answering from a knowledge base you approve and passing questions outside it to a person.
- 02
Custom models
Models trained or tuned for one task in your business, where an existing service does not do it well enough on your data.
- 03
AI API integration and microservices
AI services connected to your applications through APIs and small services around them, so a model can be changed without rewriting the application.
- 04
Computer vision and NLP
Models that read images, video and text, such as visual inspection, document and message classification and language tasks, tested on your own examples.
- 05
Deployment, monitoring and lifecycle management
Models deployed to your cloud and monitored for accuracy, cost and drift, then retrained when results slip and retired when no longer needed.
- 06
AI governance and compliance controls
Rules for what each model may see, say and do, with access control, review of high-risk answers and a log of every question and answer.
How an engagement runs.
One request path from a person to your data, drawn three times.
You can stop at any phase boundary. The assessment is free, and each phase is priced in writing before it starts.How we engage
India specifics and cost.
India specifics
- DPDP
From 13 May 2027, the Rules require monitoring and a year of logs for personal data, which prompts, chat logs and searched documents can hold. The design sets what is stored, and for how long. Legal interpretation stays with your counsel.
Source: Digital Personal Data Protection Rules, 2025 (external site)
- Data residency
Where data must stay in India, the model, the search index and the logs are placed in Microsoft Azure’s India regions. The design says which model options that allows.
- CERT-In
The Directions of 28 April 2022 require logs kept within India for a rolling 180 days. AI services and the applications around them log to such a store, with clocks synchronised to NIC or NPL.
What drives the cost
- 01
Use case: a chatbot, a staff assistant, a custom model, or vision or language work.
- 02
Knowledge sources and systems the model must reach, and how often they change.
- 03
Controls: what the model may see, say and do, and which answers a person reviews.
- 04
Usage volume and the hosting it needs, including any India-region requirement.
The technology we run for this.
Cloud platform and identity
Our own information security and privacy management systems, run under both standards.
Certifications and technologyQuestions buyers ask about this.
Related services
Why do AI pilots stall before production?
Usually because the pilot ran on clean demo data, nobody owned the exceptions, or nothing reviewed results before they reached a system of record. We start on your own documents and data. The review step and the exception path are designed before the build, and the success criteria are agreed in writing.
How do you control what a GenAI assistant can see and say?
With controls designed before the build. The assistant searches only the sources you approve, within the access of the person asking. Instructions and filters limit what it answers, and high-risk answers go to a person. Every question and answer is logged, so it can be checked later.
Should we fine-tune a model or have it search our documents?
Usually search first. An assistant that retrieves from your documents stays current when they change and can show its source. Fine-tuning suits a task the model does badly however well it is instructed, such as a house style or a narrow classification. The proof tests which one your use case needs.
What happens after the AI goes live?
It is monitored, not left to run. Accuracy, cost and drift are tracked against the criteria agreed at the proof, and the model is retrained, replaced or retired when results slip. Support windows and response times are agreed in writing before we start, and the runbooks are yours.
Can you build AI into an application we already run?
Yes. The AI is connected to your application through APIs and small services around it, so the application itself changes as little as possible. The AI part can be changed or switched off on its own. Nothing is replaced unless the assessment shows it should be, and says why.
Start with a bounded proof.
It costs nothing and commits you to nothing. You get a written list of where your IT environment is exposed, and your team can act on it with or without us.
What we put in writing
- The written findings are yours whether we ever work together or not.
- Each phase is fixed-scope and signed off separately.
- You can stop at any phase boundary.
- Runbooks, configurations and credentials are yours.