FourCents India is an AI development company in India that builds AI agents, document reading, WhatsApp assistants and workflow automation for businesses in India and abroad. We use models from OpenAI, Anthropic and the open-model world, and we wrap them in the parts that make AI dependable in real use: evaluation, fallbacks, audit logs and a person who checks what matters.
A custom AI agent or automation from FourCents India typically costs ₹2 lakh to ₹8 lakh and takes 4 to 10 weeks, excluding GST. That covers document reading, chat or WhatsApp assistants, search over your company files and automated workflows, with evaluation tests and human review built in. You own the code, the prompts and the cloud accounts.
What is an AI agent, in plain terms?
An AI agent is software that uses a language model to decide the next step in a task, then takes that step using tools you allow it to use: reading a file, searching a database, filling a form or sending a message. A chatbot answers. An agent does work.
Most useful business AI is narrower than the word suggests. It reads an invoice and fills your accounting system. It answers a customer on WhatsApp using your price list. It flags the one document in a hundred that a person needs to look at. Narrow and checked beats clever and unpredictable.
What do you get?
Each AI project is built around one job your team does today, and includes:
- The AI workflow itself, from input (a document, a message, a form) to output (data in your system, a reply, a flag for review).
- Document reading and OCR for scans, phone photos and PDFs: invoices, KYC documents, purchase orders, lab reports, contracts.
- Assistants on WhatsApp, web chat or inside your app, answering from your own content, with a hand-off to a person.
- Search over your company documents using retrieval-augmented generation, usually called RAG, with links back to the source so answers can be checked.
- An evaluation set: real examples with known correct answers, run against every change before it reaches your users.
- A review console where your team approves, corrects or rejects what the AI produced, with every decision logged.
- Cost and usage monitoring, so you see what each document or conversation costs to process.
- Launch, the first 30 days of fixes, then an optional care plan.
Who is this for?
Operations teams buried in documents. Logistics firms reading delivery challans, lenders and NBFCs checking KYC packs, clinics reading lab reports, and accounts teams entering supplier invoices into Tally.
Businesses answering the same questions all day. Coaching institutes, real estate developers, clinics and D2C brands that get hundreds of WhatsApp messages about fees, availability, order status and timings.
Companies with knowledge locked in files. Policies, SOPs, product manuals and past proposals that staff search by asking the one person who remembers. RAG search turns that into an answer with a source link.
Product teams adding AI to an app, in India, Canada, the US, the UK and the Gulf. If the product does not exist yet, see MVP development.
Off-the-shelf AI tool or a custom agent?
Many teams should start with an off-the-shelf tool. A custom build makes sense when the tool cannot see your data, cannot act in your systems, or cannot be checked.
| Off-the-shelf AI tool | Custom AI agent | |
|---|---|---|
| Time to start | Same day | 4 to 10 weeks |
| Upfront cost | Low, monthly per-seat fees | ₹2 lakh to ₹8 lakh build cost |
| Works with your systems | Through the tool's own integrations | Built into your database, Tally, CRM or app |
| Your data | Sits with the tool vendor | Stays in accounts you own |
| Accuracy checks | Whatever the vendor provides | Your own evaluation set and review rules |
| Fits your exact process | You adapt to the tool | The tool adapts to you |
| Running cost at volume | Grows with seats | Grows with usage, usually lower per task |
Our advice: if one person uses AI to draft emails, buy a subscription. If AI handles customer data, money or decisions, build it so you can control what it does.
How do we build it?
AI projects follow our standard process, with one addition: we measure accuracy before we promise anything.
- Estimate. Use the estimator at the end of this page for an itemised range in about two minutes.
- Scoping call within one business day. A 30 minute call with the engineer who will lead the build. Bring ten real examples of the work: documents, chats or tickets.
- Fixed proposal within three business days of the call. Every feature priced separately, with milestone payments tied to builds you can test.
- Two-week sprints. The first sprint usually builds the evaluation set and a first version run against it, so you see real accuracy on your own examples early. Your staging link is updated every week.
- Launch with review on. New AI workflows start with a person approving outputs. As the numbers prove themselves, you decide which cases can skip review.
- Care plan. Monitoring, model updates, prompt changes and evaluation runs, typically 15 to 20 percent of the build cost per year.
We sign an NDA on request before you share any documents, and the code, prompts, evaluation data and AI provider accounts are in your name from week one.
See your price before you call anyone
Pick your features and get an itemised estimate in two minutes.
Build my estimateEvaluation and human review
The difference between an AI demo and an AI system is evaluation. A demo works on the five examples someone tried. A system is tested on hundreds of real cases, including the ugly ones: blurred photos, handwritten notes, mixed Hindi and English, half-filled forms.
We build an evaluation set from your real examples and score every version against it. When a model or prompt changes, the scores tell us whether things got better or worse before your users find out.
Human review is designed in, not bolted on. You choose the rules: every case reviewed at first, then only cases where the AI is unsure, where amounts cross a limit, or where documents disagree. Every decision, by the AI or a person, goes into an audit log.
Data privacy and the DPDP Act
India's Digital Personal Data Protection Act, 2023 applies to any business that processes digital personal data of people in India. For AI projects the practical requirements are clear notice and consent for the purpose you collect data for, using it only for that purpose, reasonable security safeguards, breach reporting, and honouring requests to access, correct or erase data.
We build those into the system:
- Documents encrypted at rest, with access limited by role.
- Personal data purged on a schedule you set in your privacy policy.
- Only the fields a model needs are sent to it, and identifiers are masked where possible.
- AI provider accounts in your name, on business API terms. OpenAI and Anthropic both state that data sent through their business APIs is not used to train their models by default. Check their current terms, and we will walk you through the settings.
- Where data must not leave your servers at all, an open model hosted in your own cloud account.
This is engineering, not legal advice. For a large or sensitive dataset, have your lawyer review the notice and consent text.
Which AI model should you use?
We are not tied to one provider. We pick per task, and we often use more than one in the same system.
- OpenAI models are strong all-rounders for chat, extraction and voice, with mature tooling.
- Anthropic's Claude models are strong at long documents, careful instruction following and writing.
- Open models such as Llama, Qwen or Mistral can run on your own servers. They cost more to set up and host, but no data leaves your account and there is no per-request fee.
Because the evaluation set exists, switching models is a measured decision, not a guess. When a cheaper or better model is released, we run it against your examples and switch only if it wins.
What does AI cost to run?
Running costs are mostly per-use fees to the AI provider, charged by the volume of text, images or audio processed. Prices fall every year and differ widely between models, so we estimate yours from your real volumes and current published rates on OpenAI's pricing page and Anthropic's pricing page, rather than quoting a figure here that will be out of date.
We keep the bill down by using the cheapest model that passes your evaluation set, caching repeated instructions and routing only hard cases to a stronger model. You see the cost per document or conversation in the dashboard.
How much does AI development cost in India?
An AI agent or automation from FourCents India costs ₹2 lakh to ₹8 lakh, excluding GST, and takes 4 to 10 weeks. Indian invoices carry 18 percent GST.
What moves the price:
- How many steps the agent takes. Reading one document type is simpler than reading ten and reconciling them against each other.
- Integrations. WhatsApp, a CRM, Tally or an ERP add ₹60,000 to ₹2 lakh, because you depend on someone else's system.
- Review tooling. An admin dashboard for reviewers and reports adds ₹80,000 to ₹2.5 lakh.
- Languages. Hindi and regional languages add ₹40,000 to ₹1.5 lakh.
- Accuracy target. Getting from good to very good on messy inputs takes more evaluation work.
If you are adding AI to an existing app, the AI features add-on (reading, chat, search) costs ₹1.2 lakh to ₹4 lakh. See every line, including a worked example of a WhatsApp assistant that reads invoices, on our pricing page.
Proof: what we have built with AI
VisaVisa is our strongest example of AI and automation working together. It prepares US, UK and Schengen visitor visa applications for Indian travellers. Uploaded documents are encrypted at rest and read automatically. AI checks flag unreadable, expired or inconsistent documents before an expert sees them, and software cross-checks names, dates and passport numbers against the forms. A scheduler moves each case forward, retrying document reads, sending reminders and alerting the team when something needs a person. Every case gets a reviewer check and a senior sign-off, and it includes a refusal-letter reader.
SPAIKING is a voice-first AI tutor for speaking a new language. Our own server relays the realtime voice session, so the AI provider key never sits on the phone, and the learner can interrupt the tutor mid-sentence.
PaperQuant takes an options strategy described in plain English, turns it into a structured strategy, and backtests it on simulated prices. See all our work.
Questions clients ask
How much does it cost to build an AI agent in India?
A custom AI agent or automation from FourCents India costs ₹2 lakh to ₹8 lakh, excluding GST. A single document type with human review sits at the lower end. Multi-step agents with integrations, regional languages and detailed review tools sit higher. Running costs for the AI provider are separate and depend on your volume.
Can AI read scanned documents and handwriting?
Modern models read printed scans and phone photos well, and handwriting reasonably well. Accuracy depends on your documents, so we measure it on your real examples in the first sprint before you commit to the rest of the build.
Is my data safe with OpenAI or Anthropic?
Both providers state that data sent through their business APIs is not used to train their models by default, and the accounts are set up in your name. For data that must never leave your servers, we can run an open model in your own cloud account instead.
Does the DPDP Act stop us from using AI on customer data?
No. The DPDP Act, 2023 requires notice, consent for a stated purpose, security safeguards and respect for people's rights over their data. We build those controls in, and suggest your lawyer reviews the notice text for sensitive data.
Can you build a WhatsApp assistant for our business?
Yes. We build assistants on the official WhatsApp Business Platform that answer from your own content, take bookings or orders, and hand the chat to a person when needed. WhatsApp's own messaging charges are paid by you to Meta or your provider.
What is RAG and do we need it?
RAG, retrieval-augmented generation, means the AI first searches your documents and then answers using only what it found, with links to the source. You need it when answers must come from your policies, manuals or records rather than the model's general knowledge.
What happens when the AI gets something wrong?
It will, sometimes, which is why we design for it. Uncertain cases go to a person, every action is logged, and each correction can be added to the evaluation set so the same mistake is caught in future versions.
Can you add AI to software we already have?
Yes. We connect to your existing app, database or tools through their APIs. If the software is old or undocumented, we start with a short paid audit before quoting.
Get your estimate
Pick the features you need. The receipt uses the same prices as our pricing page, and you can send it for a fixed proposal. We reply within one business day, or you can contact us directly. Related services: custom software development, mobile app development, web development and ecommerce development.