TL;DR: Jupiter Astro-AI is not “AI sprinkled on horoscopes.” We use deterministic sidereal astronomy for chart math, large language models for explanation and content, and generative image tools for marketing assets — with humans reviewing product decisions at each milestone.
This article explains how the product is put together for anyone asking: “Where is AI actually used, and where is real math?”
┌─────────────────────────────────────────────────────────────────────────┐
│ USER (browser / PWA) │
└─────────────────────────────────┬───────────────────────────────────────┘
│
┌───────────────────────┼───────────────────────┐
▼ ▼ ▼
┌──────────────┐ ┌──────────────┐ ┌──────────────┐
│ Web app │ │ Secure │ │ Payments │
│ (React) │ │ sign-in │ │ & tokens │
└──────┬───────┘ └──────────────┘ └──────────────┘
│
▼
┌──────────────────────────────────────────────────────────────┐
│ Application API (hosted) │
│ ┌────────────────┐ ┌─────────────────┐ ┌─────────────┐ │
│ │ Chart engine │ │ AI readings & │ │ Weekly blog │ │
│ │ sidereal math │ │ forecasts │ │ heroes & │ │
│ │ Lahiri ayanamsa│ │ muhurt timing │ │ editorial │ │
│ └────────────────┘ └─────────────────┘ └─────────────┘ │
│ │ │ │ │
│ └────────────────────┴────────────────────┘ │
│ ▼ │
│ Database (profiles, saved readings) │
└──────────────────────────────────────────────────────────────┘
Key principle: Positions and houses are computed in code. Language, narrative, and images are AI-assisted.
| Layer | AI role | Non-AI backbone |
|---|---|---|
| Product & UX | Design iteration, copy drafts | Human product approval |
| Kundli chart | AI explains chart in readings | Sidereal ephemeris + Lahiri ayanamsa |
| Weekly map / transits | AI interprets transits for your houses | Ephemeris longitudes, house mapping |
| Full reading | AI narrative from structured chart data | Chart facts + dasha context |
| Muhurt finder | AI describes windows in plain English | Date-range search + chart rules |
| Share captions | AI light-hearted social snippets | Reading text you already generated |
| Weekly Rashi blog | AI writes 12 signs + headline | Generated hero from brand assets |
| Support | Templated contact delivery | Spam protection on forms |
| Infra | Documented deploy runbooks | Standard hosting, database, HTTPS |
We do not ask a language model to invent planetary longitudes. That would fail users and fail any serious review.
Vedic apps live or die on ayanamsa and time. Our server computes:
For birth instant (t) (UTC), Julian Day (JD):
[
JD = \frac{\text{Unix ms}}{86400000} + 2440587.5
]
Lahiri (Chitrapaksha) ayanamsa in degrees (linear model used in production):
[
\psi(JD) = 23.85675 + 1.39642 \cdot \frac{JD - 2451545}{36525}
]
Tropical ecliptic longitude (\lambda_{\text{trop}}) from a high-precision ephemeris:
[
\lambda_{\text{sid}} = (\lambda_{\text{trop}} - \psi(JD)) \bmod 360^\circ
]
[
\text{sign index} = \left\lfloor \frac{\lambda_{\text{sid}}}{30} \right\rfloor,\quad
\text{degree in sign} = \lambda_{\text{sid}} \bmod 30
]
Ascendant: sidereal RAMC formula from local sidereal time and latitude — not guessed by AI.
Whole-sign house of planet (P):
[
\text{house}(P) = ((\text{sign}(P) - \text{sign}(\text{Lagna})) \bmod 12) + 1
]
Mean north node tropical longitude (polynomial in Julian centuries (T)):
[
\Omega = 125.044522 - 1934.136261,T + 0.0020708,T^2 + \frac{T^3}{450000}
]
Then ( \lambda_{\text{Rahu}} = \Omega - \psi ), ( \lambda_{\text{Ketu}} = \lambda_{\text{Rahu}} + 180^\circ ).
Performance: A full chart for seven grahas plus nodes runs in milliseconds on our servers — the “ton of maths in seconds” claim on our homepage is literal timing, not marketing fluff.
Birth datetime + place
│
▼
┌──────────────┐ Structured chart facts
│ Chart engine │─────────────────────────┐
│ (deterministic) │
└──────────────┘ ▼
┌──────────────┐
│ AI narrative │
│ + safety │
│ guardrails │
└───────┬──────┘
▼
Plain-English reading
saved to your account
Guardrails (non-AI): sign-in required, token wallet, rate limits, and clear disclaimers — so usage scales with real accounts, not anonymous abuse.
| Phase | Human | AI-assisted |
|---|---|---|
| Ideation | Problem: generic horoscopes useless for decisions | Feature lists, competitive framing |
| Chart engine | Validate formulas, pick Lahiri | Boilerplate, tests, refactors |
| API design | Tiers, auth policy | Route scaffolding, middleware patterns |
| Frontend | Brand colours, UX priorities | Components, layout, copy drafts |
| Weekly blog system | Editorial tone, 12-rashi structure | Structured generation, image briefs |
| Operations | Domains, secrets, monitoring | Deploy checklists, runbooks |
| Blog & SEO | Series strategy | Long-form article drafts |
Every change is reviewed in version control before release — the same discipline as any production engineering team.
Scheduled or manual publish
│
▼
AI → headline, 12 rashis, dos/don'ts, highlights
│
▼
Generated hero image (brand-aligned layout)
│
▼
Published edition → /blog/weekly/:slug
This is generative media, clearly labelled as AI-assisted editorial — not presented as a human pundit’s handwritten palm leaf.
Built with: modern web stack, secure auth, token-based AI features, precise ephemeris math, and disciplined engineering.
Questions about how the product works? Use our Contact page — we are happy to explain the approach without sharing internal credentials or vendor contracts.