AI at Vedatricks
AI where it ships, not where it sells.
One customer-facing copilot is live today, in HamaraPG. Everything else is listed below with the status it really has — in build, internal, or a prototype.
The inventory
What runs where.
| Product | What it does | 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 | Launching soon |
| 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 |
Two things this table deliberately does not say. Biztricks’ Smart Insights cards are a rule-based engine, not AI, and we do not market them as AI. And there is no Vedatricks product where voice AI is in production — the voice work is a prototype.
Providers
Who processes the words.
- Live
OpenAI
Runs what is live today: the HamaraPG owner copilot and the internal VedaERP assistant, both on gpt-4o-mini with function calling, read-only queries and per-organisation token caps.
- In build
Sarvam AI
The planned provider for the assistants in build and for this site’s assistant, chosen for Indian-language conversation and Indian data handling. Sarvam holds ISO 27001 and SOC 2 Type II certification. Vedatricks itself holds no ISO or SOC certification and does not claim one.
- Prototype
Speech, in prototype
Ovi, our voice intake prototype, uses Sarvam speech models for Indic transcription and synthesis. It is not in production, it is not sold, and no clinic runs on it.
Guardrails
The same rules, in every assistant.
These are the rules this site’s assistant enforces, and the specification every assistant we build next is held to. The one already in production, HamaraPG’s copilot, enforces its own set: a read-only query registry, four tool rounds and a twelve-message window per conversation, a monthly token cap per organisation, ten requests a minute, and every turn written to an audit table.
-
Read-only by default
Product assistants answer from a fixed set of organisation-scoped queries written by us. The model chooses which question to run; it never writes SQL, and it never changes your data.
-
Scoped, and it says no
The website assistant answers about Vedatricks products, pricing, careers and support. Anything outside that scope gets a refusal and an offer to fetch a person.
-
No professional advice
No medical, clinical, legal, financial or tax advice, from any assistant we ship. Clinical questions are refused and handed to a qualified human.
-
Identifiers are stripped first
This site’s assistant redacts Aadhaar, PAN, card, phone and email patterns from a message before it reaches a model provider. Every new assistant is built to the same rule.
-
Hard caps on use and cost
Turn limits per session, request limits per user and per IP, a monthly token cap per organisation and a daily spend ceiling checked before every call.
-
Every turn is logged
Session, model, tokens and cost are recorded for each turn, and product copilots write to an audit trail the account owner can inspect.
-
A person, one click away
Any conversation can be handed to our team, and repeated refusals hand it over automatically. The assistant is a front desk, not a wall.
-
A kill switch we actually use
Each assistant reads an enable flag on every request. If something is wrong we turn it off from the admin panel and the interface says so.
Disclosure and data
What you are told, and what we keep.
Assistant replies are labelled as AI-generated, and every assistant carries a standing note that it can be wrong. You are told you are talking to software before your first message, not after it.
- Conversation retention
- When the site assistant goes live, transcripts will be kept for 90 days and then deleted. Access logs are kept for one year, as the DPDP Rules require.
- Who else sees a message
- The model provider for that assistant — OpenAI for the HamaraPG copilot, Sarvam AI for what is in build — and no one else. Providers may process outside India; the table above says which provider sits behind which assistant, and Sarvam AI is listed as a processor in our Privacy Policy.
- Model training
- We do not want your conversations used to train models, and we say so to every provider we sign with. Where a provider's terms or our contract with them settle the point, we will name the provider and the term in our AI Disclosure rather than summarise it here. Read the AI Disclosure.
- What not to send
- Do not put Aadhaar numbers, card details, passwords or health records into any assistant. Identifiers are redacted before a message leaves us, but the simplest protection is not to type them.
- Your rights
- Ask for access, correction or erasure of anything we hold, including transcripts, through our privacy request form. We answer within 30 days, and always within the 90-day statutory limit.
Roadmap
What is next, without dates.
We publish no launch dates for AI features. A capability moves to “live” on this page on the day it is live in the product, and not one day earlier.
- HamaraPG Live
- VedaERP (internal) Internal
- Clinitricks Launching soon
- Biztricks In build
- Ovi (voice intake) Prototype
- Veda (website assistant) In build
This website’s assistant
Veda, the assistant on this site.
Veda answers questions about Vedatricks products, pricing pages, careers and support in English, Hindi, Kannada and Marathi, and hands you to a person whenever you ask. It is in the corner of this page now; the model behind it is still being brought into service, so the dock says so plainly when it cannot answer. Either way, the first thing it does is tell you it is an AI, name its provider and ask you to continue before anything is sent.
- It answers about Vedatricks only, and refuses everything else.
- It gives no medical, legal, financial or tax advice.
- It links only to pages on this site and to our own product sites.
- A challenge check runs when a conversation starts, to keep automated abuse out.
- Every reply is labelled as AI-generated and can be wrong; “Talk to a human” is always available.
Planned provider: Sarvam AI. This page is updated when that changes.
Ask us what our AI actually does.
We would rather answer the awkward version of that question than publish a claim we cannot back.