The technology

The engine behind Mercury & Jupiter.

AstroCapitalX is not a chatbot with an astrology prompt. Mercury & Jupiter are a proprietary AI system: our own astronomy compute engine and a transparent research pipeline ground every answer in real planetary mathematics — not vibes, and not hallucination.

≤0.005°Engine calibration vs Drik Panchang
9 · 14Planets computed · Sectors modelled
0Third-party astrology APIs
2Domain AI models (Mercury · Jupiter)
01
Compute · proprietary

The sidereal astronomy engine

At the core is our own self-hosted ephemeris engine — no third-party astrology API. It computes true sidereal (Lahiri ayanamsa) planetary longitudes and speeds, Vimshottari dashas, transits, and location-aware panchang, sampled at IST day boundaries with a binary-search station finder for ~1-minute precision. It is calibrated to ≤0.005° against Drik Panchang and runs a boot-time smoke test so a wrong ayanamsa can never ship silently.

  • Real astronomy, MIT-licensed, in-process — no external dependency
  • True sunrise/sunset per location (equation of time + refraction)
  • Dashas & transits as structured data, not text
  • Eclipse & muhurat computation from first principles
02
Validate · research

The research & backtesting engine

Signals are not asserted — they are tested against chance. We cast first-trade charts for pilot assets and run each candidate indicator through a rigorous quant pipeline: Deflated Sharpe Ratio, walk-forward / out-of-sample validation, and Probability of Backtest Overfitting (PBO) via CSCV. A signal is only called "beats chance" when DSR ≥ 0.95, PBO ≤ 0.5 and OOS > 0. Where the evidence is weak, we say so — publicly.

  • Deflated Sharpe to correct for multiple testing
  • Walk-forward OOS so nothing is fit in hindsight
  • PBO / CSCV to quantify overfitting risk
  • Indicators tagged evidenced · tradition · folklore
03
Ground · retrieval

The domain knowledge base

Mercury & Jupiter draw on a curated Vedic financial-astrology corpus — a human-reviewed knowledge base of doctrine, yogas, dasha effects and sector rulerships. Relevant passages are retrieved per question (BM25 lexical retrieval, top matches) and injected into the model's context, so answers are anchored to real doctrine rather than improvised. Only vetted, approved entries reach the retrieval layer.

04
Reason · inference pipeline

The Mercury & Jupiter inference pipeline

Each model is a layered inference pipeline, not a single prompt. Every query is assembled from: a base persona, a domain overlay (Mercury for markets, Jupiter for business & life), the retrieved knowledge, the user's live computed chart, and today's real transit & panchang context — then run through a temporal-safety layer (the model may never quote a past price as current) before a large language model generates the response. Output is post-processed for plain-text, multilingual delivery.

  • Chart-grounded — your real kundli, computed live
  • Transit-aware — today's actual sky, every reply
  • Temporal-safe — never fabricates prices or dates
  • Multilingual — answers in your language

How a single answer is generated

Input
Your question
+ your saved birth chart
Compute
Engine runs
live planets, dashas, transits
Retrieve
Knowledge base
matching Vedic doctrine
Ground
Context assembled
persona + chart + sky + doctrine
Reason
Model + safety
LLM + temporal guardrails
Output
Grounded answer
in plain language

What we don't claim

We build on a foundation large language model for language and reasoning — the proprietary work is the astronomy engine, the backtested research methodology, the curated corpus and the grounding pipeline that surround it. AstroCapitalX is educational and is not SEBI-registered; our outputs are a Vedic lens on timing and sentiment, never buy/sell recommendations or financial advice.