Aavyalabs Logo
Home/Use Cases/Aadikarta — AI-Powered Astrology Consultation Platform
Marketplace & Consumer Tech

Aadikarta — AI-Powered Astrology Consultation Platform

Live expert marketplaces fail on three things: billing that can be gamed, sessions that handle drops unfairly, and matching that sends the wrong expert to the wrong user. We engineered all three — a tamper-proof server-authoritative billing engine, a 7-state session state machine, and an ML matching layer that improves with every consultation.

MarketplaceAI/MLReal-TimeWebSocket
₹0
Billing disputes in UAT
7-state
FSM covering every edge case
3
Billing models supported
10+
AI/ML features across 4 layers

The Challenge

The client needed a live consultation marketplace where billing couldn't be gamed, sessions survived network failures fairly, and the right astrologer reached the right user — none of which off-the-shelf solutions could deliver.

  • Server-authoritative per-minute billing resistant to manipulation
  • Session state that survives network failures equitably
  • Astrologer matching beyond basic availability algorithms
  • A personalisation layer to drive retention
  • Full auditability — everything engineered from first principles

Our Solution

Aavyalabs built a tamper-proof, server-authoritative billing engine, a 7-state session finite state machine handling every edge case from network loss to mid-session recharge, and an ML-powered matching layer that improves with every consultation.

  • Per-minute, fixed, and hybrid billing — server-authoritative, zero disputes in UAT
  • 7-state session FSM with grace periods and pro-rated refunds
  • ML matching on specialty, language, performance, and availability
  • AI personalisation: concern classification, Kundli briefs, churn prediction

Key Features & Capabilities

What we designed, built, and shipped.

Real-Time Billing

Per-minute, fixed, and hybrid models; server-authoritative; zero disputes in UAT.

Session Management

7-state FSM covering edge cases, grace periods, and pro-rated refunds.

ML Matching

Considers specialty, language, past performance, and availability.

Concern Classification

NLP auto-tagging from the user's first message.

Kundli Analysis

Birth data → structured Jyotish interpretation in seconds.

Consultation Summaries

GPT-4 PDF generation delivered after every session.

Churn Prediction

Behavioural modelling with push-notification triggers.

Fraud Detection

Isolation Forest anomaly detection on transactions.

Delivery Roadmap

1

Session Lifecycle (FSM)

REQUESTED → ACCEPTED → ACTIVE → PAUSED → RESUMED → ENDED → REFUNDED

Technology Stack

PythonFastAPIWebSocketsCeleryRedisPostgreSQLOpenAI GPT-4spaCyscikit-learnProphetpgvectorIsolation ForestReact NativeNext.jsRazorpayAWS S3Firebase PushMiroTalk

Building a Live Expert Marketplace?

Per-minute billing, fair session handling, and smart matching are hard to get right. We've engineered all three — tell us about your marketplace and we'll map the architecture.