Vedatricks
Book a 30-minute call

AI integration

AI you can put in front of people who pay you.

Read-only assistants over your own data, with hand-written safe queries, audited turns and hard spend caps — the pattern already running in HamaraPG.

Businesses with an existing system and questions their team keeps asking it.

Start a project All services

What you get

Everything below is in the scope, in writing.

Read-only by default
The model picks from hand-written queries; it never writes SQL.
Confirm before commit
Any action that changes data is shown and confirmed by a person.
Scoped to one organisation
Every query carries the tenant scope; cross-tenant reads fail closed.
Audited turns
Prompt, answer, model and tokens stored for every conversation.
Hard caps
Per-user rate limits, per-organisation token caps and a global spend cap.
PII redaction
Identifiers stripped before text leaves your systems for a model.
Indian languages
Hindi, Hinglish and English in production today; more via Indic models.
A human handoff
Two refusals or one request and the conversation reaches a person.

AI at Vedatricks

What runs where, stated plainly.

We will not sell you a capability we have not built. This is the whole list, with its real status.

AI capabilities across Vedatricks products and their status
Product Capability Provider Status
HamaraPG Owner copilot: Hindi/Hinglish/English Q&A over dues, defaulters, occupancy, P&L, complaints, compliance, vacancies (read-only, audited, token-capped). OpenAI (gpt-4o-mini) Live
VedaERP (internal) Assistant that turns natural language into CRM actions with user confirmation. OpenAI (gpt-4o-mini) Internal
Clinitricks AI Assistant v1: read-only questions over appointments, follow-ups, revenue; voice-to-prescription drafts with doctor approval. Planned: Sarvam 105B + Saaras STT Internal
Biztricks Quote/lead copilot over pipeline and receivables; HSN suggester. Planned: Sarvam In build
Ovi (voice intake) AI voice patient intake in Indic languages. Sarvam speech + LLM NLU Prototype
Veda (website assistant) This site’s assistant dock: answers about Vedatricks products, pricing pages, careers and support in EN/HI/KN/MR; hands off to a human on request. Sarvam AI (sarvam-105b-conversations) In build

Guardrails

The safety model ships with the feature.

These are not aspirations. They are the controls we build into every assistant, taken from the pattern already running in production.

  • Scoped system prompt — the assistant answers about your business and refuses everything else.
  • Read-only query registry — the model chooses a hand-written query; it never writes SQL.
  • Organisation scoping that fails closed, with tests to prove it.
  • Identifier redaction before any text is sent to a model provider.
  • Rate limits per user, token caps per organisation and a global spend cap with an 80% alert.
  • Every turn stored with model, tokens and cost; transcripts on a stated retention period.
  • An "AI-generated" label on every reply and a one-click handoff to a person.
  • A kill switch an administrator can use without a deploy.

How we work

Five steps, no surprises.

The same process for a ₹3,000 site and a multi-tenant platform. Only the size of each step changes.

  1. Scope call

    One call to understand the work, the people who will use it and the deadline. We reply during office hours.

  2. Written scope and price

    A fixed written scope, a price and what is explicitly out of scope. Nothing starts before you sign it.

  3. Build in visible slices

    You get a working URL early and a slice you can click every week, not a status report.

  4. Handover

    Source, credentials, documentation and a walkthrough with your team. The accounts are yours.

  5. Support

    An agreed support window after launch, then a maintenance arrangement if you want one.

Proof

Ask it in Hinglish who has not paid. It answers from your own records.

The HamaraPG owner copilot answers questions about dues, occupancy, complaints and monthly P&L for one organisation at a time — read-only, audited and budget-capped. Where an assistant is still being built, its own page says so.

Live

Ask the copilot: "kaun kaun defaulter hai?"

The owner asks in Hinglish; the copilot answers from audited, org-scoped queries over dues, defaulters, occupancy, P&L, complaints and vacancies.

  1. Owner types in Hinglish
  2. Copilot picks a safe, read-only query
  3. Answer with names and amounts
  4. Every turn audited and token-capped

Provider: OpenAI Chat Completions (gpt-4o-mini, configurable). Languages: Hindi, Hinglish, English.

Read the HamaraPG product page →

What we do not claim

  • No clinical, legal or financial advice from any assistant we ship.
  • Voice assistants are a prototype at Vedatricks, not a production capability.

Questions

Asked before every project.

What AI have you actually shipped?

One customer-facing copilot: the HamaraPG owner assistant, which answers questions about dues, defaulters, occupancy, P&L, complaints, compliance and vacancies in Hindi, Hinglish or English. It is read-only, organisation-scoped, audited and token-capped. Our internal ERP has a second assistant. Everything else on this site is labelled in build or prototype.

Will my data be used to train a model?

Not by us, and our position is that no provider should either. We send the minimum text needed, redact identifiers first, and ask each provider for its no-training terms or opt-out before we sign. The provider, its training terms, the retention period and the processing region are written into your agreement before we build.

Which model provider do you use?

It depends on the job. Our shipped copilot runs on OpenAI. For Indic-language work we build on Sarvam AI, which is India-based and ISO 27001 and SOC 2 certified as a provider. We will tell you what runs where rather than hiding it behind the word "AI".

What stops it inventing an answer?

Scope. The assistant answers from a fixed set of queries against your own data, refuses topics outside its brief, and hands off to a person on request or after repeated refusals. It carries an "AI-generated" label, and it gives no medical, legal or financial advice.

What does it cost to run?

Model usage is metered per token, so we set a monthly cap per organisation and a global spend cap before launch, and alert at 80% of it. You see the usage; there is no open-ended bill.

Start with one question your team keeps asking.

The best first assistant answers a question people already ask ten times a week. Bring that question to the scope call.

Start a project See a system we built
Not ready for a call?

Cookie settings

Choose what you allow. You can change this at any time from “Cookie settings” in the footer, and withdrawing is as easy as allowing.

  • Remembers choices you make on the site, such as your preferred language for the website assistant.

  • Vedatricks Analytics is first party and cookieless by default: it counts page views without writing an identifier to your device. A cookie is only set if you opt in, and only to avoid counting the same visit twice.

  • Strictly necessary Always on

    Needed for the site to work: to remember your consent choice, to protect forms from automated abuse and to keep you signed in to a Vedatricks product. These are not used for advertising and cannot be switched off from the banner.

  • Payments Always on, but only on checkout pages

    Razorpay Checkout sets its own cookies when it loads, which happens only on pages where you are actually paying. Razorpay is the controller of those cookies; see Razorpay’s privacy policy.

Full detail of every cookie is in the Cookie Policy register.