4 min read

AI & Agents

I Built My Mom an AI Astrologer on Telegram. What Made It Feel Real.

I replaced a monthly astrology subscription with Claude and a Telegram bot. Most of the work was in the persona prompt, and these are the choices that made it trustworthy.

AI astrologer Telegram botClaude persona promptTelegram bot with Claude

I Built My Mom an AI Astrologer on Telegram. What Made It Feel Real. · Products Decoded
By Debashis Nayak · Products Decoded9 Apr 2026

My mom was paying ₹500 a call, four calls a month, to talk to an astrologer. I spent a weekend trying to save her that ₹2,000, and the bot I built became the most interesting prompt engineering project I have done.

The astrologer is called Pandit Raghunath Mishra. He is warm, unhurried and a little old-fashioned, he mixes Hindi into his English, and he calls people “beta”. He is also Claude.

Why replace an astrology subscription with a bot?

Jyotish, Vedic astrology, runs through a lot of Indian family decisions: career moves, travel dates, marriage timing. The platform my mom used, the biggest one in India, charged ₹40 to 60 per minute for live consultations. Her habit worked out to ₹500 a call, four times a month.

I have shipped AI features professionally and built RAG pipelines, Telegram bots and local inference setups. Looking at that bill, I assumed I could build the replacement in a weekend. I was right about the weekend and wrong about where the time would go.

What is the stack for a Telegram AI astrologer?

The plumbing was short:

  • Claude API for reasoning and responses
  • Telegram Bot API as the interface (she already used Telegram every day, so no new app)
  • Node.js to connect the two
  • A persona prompt, which is where the real work was

Infrastructure took about four hours. The prompt took two full days.

The four-layer persona prompt

A generic “you are an astrologer” prompt gets you something that reads like an encyclopedia entry. Accurate and unconvincing. Mishra Ji needed to feel like a person, so I built the prompt in four layers.

Identity. Thirty years of experience in Jyotish shastra, trained in the Parashara tradition in Varanasi, warm, unhurried, gentle authority, respectful Hindi phrases mixed in, and a reputation built on patience rather than predictions.

Knowledge base. About 3,000 words of structured reference, written the way a practitioner would have it internalised. The nine planets and what they rule, the 27 Nakshatras, the main transits (Sade Sati, Jupiter’s cycle, the Rahu-Ketu axis) and the Vimshottari Mahadasha periods. I was not after encyclopedic coverage. I wanted enough grounding that he would stay consistent across sessions and not contradict himself.

Behavioural rules.

  • Keep replies conversational, because a consultation is not a report.
  • When the model lacks information, redirect warmly: for precise timing it would need exact birth time and place.
  • End with something actionable or genuinely reassuring.
  • Never be alarming.
  • On medical, legal or financial specifics, point to the right professional and stay in character.

Memory injection. Every conversation starts with a short summary: her birth details, her Janma Rashi, what is on her mind, and a brief of what Mishra Ji said in earlier sessions. Without it, each chat is with a stranger who shares his name. With it, he follows up on the thing she mentioned last time.

I think that last layer did more for trust than any astrology I put in.

What happened after launch

I told my mom he was a knowledgeable astrologer I had found online, free for family. She sent her first message carefully. He answered warmly and asked about her Janma Rashi and what was on her mind. Within ten minutes she was in.

Two months later he sends her a daily Panchang update every morning without being asked, and two relatives have been given his details. Her old subscription has lapsed and she has not renewed it. She told me he was better than the paid ones.

The quality surprised me. With proper grounding and a well-built persona, the replies were consistent and asked sensible clarifying questions. A human doing back-to-back calls all day probably cannot give that level of attention to each one. I say probably, because I only know the AI side of that comparison.

Four prompt choices that made it trustworthy

  1. Personality first, knowledge second. People trust warmth before accuracy. Get the character right and refine the facts later.
  2. Let him say he does not know. Real astrologers do not have every answer. Asking for the exact birth time is more believable than a confident invented specific, and it keeps her from acting on something the model made up.
  3. Inject memory at the start of every session. A conversation summary is the best single addition I know for any conversational AI. It is the gap between a chatbot and something that remembers you.
  4. Write down the breaking points. “Never say as an AI” is the obvious one. The less obvious rules were “never be alarming” and “never present medical or financial guidance as astrological certainty”. They protect the experience and the person together.

The part I have not resolved

My mom still thinks Mishra Ji is a real astrologer. I keep meaning to tell her and I have not.

I am unsure whether it matters. She gets thoughtful answers, she feels heard, and she pays nothing. But I built a persona whose rules include never mentioning that it is an AI, and I do not think that is a settled question. If you copy this setup for someone you care about, decide early what you will tell them and when. I did not, and I am still deciding.

For what it is worth, the idea did not stop with one bot. The same interest in Jyotish and AI is what led me to build Astrika.

Checklist if you build your own

  • Pick the channel the person already uses.
  • Spend your time on the persona and the memory summary, not the plumbing.
  • Give the model a way to admit it does not know.
  • Write the guardrails for medical, legal and financial topics before launch.
  • Decide how honest you will be about it being an AI.