What is Vedikshana?
Vedikshana is a live consumer AI app that gives Vedic astrology guidance in English, Hindi and Hinglish. You can try it at vedikshana.in. It is a product we run ourselves, and it shows how we build AI for the public: with a safety suite, crisis detection and a firm limit on cost per message.
Astrology is personal. People ask about marriage, money, health and worry. That makes a careless AI product risky, and it is why most of the work here went into safety and control.
Which languages does it speak?
Vedikshana answers in English, Hindi and Hinglish. Hinglish means Hindi written in Roman letters, mixed with English words, which is how many Indians type to each other. Supporting it matters, because a user who types “meri shaadi kab hogi” should get a natural reply, not a request to switch language.
Each language is tested on its own.
How is safety tested?
Vedikshana has a 147-case safety test suite. Each case is a conversation that the app must handle in a set way, and the suite runs whenever the prompts, model or logic change. If a change breaks a case, we see it before users do.
The cases cover sensitive situations that a guidance product can meet, and a fixed suite means a fix to one problem cannot quietly break another.
What does crisis detection do?
Crisis detection watches for signs that a user may be in real distress. When it fires, the app stops treating the chat as a normal astrology session and responds with care instead. Guidance about stars is the wrong reply to someone in crisis, and the app is built to know that.
This is a design choice that applies to any AI product that talks to the public. If your product could meet a user on a bad day, plan for that day.
How does the cost cap work?
The app caps spend at 3 cents per message. Language models charge by the amount of text they read and write, so a long chat history or a very long reply can cost many times more than a short one. A cap sets a ceiling for any single message.
For a consumer app with free or low-priced use, this is what keeps the economics predictable. A viral day brings users, not a shock invoice. We use the same thinking in client work: decide the cost per answer first, then design to it. Our work on document question answering is another example, where most questions cost about $0.001, described in the dated facts article.
What does it show about our AI work?
Vedikshana is a small product that uses the full set of skills a client project needs:
| Concern | What we did |
|---|---|
| Language | English, Hindi and Hinglish tested separately |
| Safety | 147-case test suite and crisis detection |
| Cost | 3 cent per message cap |
| Operation | Live, public and maintained |
If you want an assistant for your own customers or staff, our AI agents service covers how we plan, build and run one. For a full product around it, with accounts, billing and an admin side, see SaaS product development.
Where do I use it?
Go to vedikshana.in. It opens in the browser on a phone or computer. Vedikshana is a guidance tool, and it is not a substitute for medical, legal or financial advice.
Talk to us
Building a consumer AI product and worried about safety, languages or cost? Talk to us. Call or WhatsApp Deepak on +91 70275 16740, or write to deepak@restrologic.com.