Code is the Astrologer.
AI is the Messenger.

The fundamental problem with most AI astrology tools is a sequencing error. They give the AI raw birth data and ask it to produce astrological insights in a single step. This forces the language model to simultaneously perform astronomical calculations (which it was never designed to do accurately) and generate empathetic, contextual narratives. The result is predictable: hallucinated planetary positions, invented yoga combinations, and confidently stated dasha periods that don't match any actual calculation. The AI isn't lying — it's doing its best to fulfill a request that requires mathematical precision with a tool built for language.

Keshoo's architecture solves this with a strict separation of concerns that we call the Pre-Synthesis model. The calculation phase and the synthesis phase are entirely separate systems running on different infrastructure. The calculation engine — built in Node.js with the Swiss Ephemeris C library — handles every mathematical operation: converting birth coordinates to topocentric positions, applying the Chitrapaksha (Lahiri) Ayanamsa correction, computing sidereal planetary longitudes to sub-arc-second precision, mapping planets to houses using whole-sign and Bhava Madhya calculations, deriving Vimshottari Dasha periods from the Moon's Nakshatra position, computing Shadbala (six-fold strength) for every planet, and evaluating Yogas and Doshas against hardcoded classical rule sets.

Only after this computation is complete — and only with the structured output data — does the AI layer receive anything. The AI model receives a detailed data package: planetary positions with house and sign placements, active dasha and bhukti periods with exact date ranges, Shadbala scores as numerical values, a list of identified yogas with their constituent planets and conditions, dosha evaluations with severity assessments, and current transit positions relative to natal placements. The AI's job is exclusively narrative synthesis: taking this verified, structured data and presenting it as clear, contextual, actionable insight in natural language. It cannot add information that isn't in the data package. It cannot override a computed house placement with a guessed one. The math acts as an unbreakable boundary.

Ayanamsa selection is one of the most consequential decisions in any Vedic astrology system. The Ayanamsa accounts for the precession of the equinoxes — the slow wobble of Earth's axis that causes the tropical and sidereal zodiacs to drift apart over centuries. Different Ayanamsa values can shift planetary sign placements by a degree or more, which is enough to change a planet's Nakshatra, its house placement near a sign boundary, or the activation date of a dasha period. Keshoo uses the Chitrapaksha (Lahiri) Ayanamsa — the standard adopted by the Indian government's Calendar Reform Committee and widely used in professional Jyotish practice and academic research. It is set so that the star Spica (Chitra) aligns with the zero point of Libra, providing a stable, historically consistent reference.

For positional calculations, Keshoo uses topocentric coordinates rather than geocentric ones. The difference is subtle but meaningful: geocentric calculations treat the observer as if they're at the center of the Earth, while topocentric calculations account for the observer's actual position on the Earth's surface. For the Moon — which is close enough for this parallax to matter — the difference can be up to one degree of arc, enough to shift Nakshatra boundaries and therefore Dasha start dates. Since Dasha timing is central to most career, relationship, and life-transition questions, this precision has real interpretive consequences.

The calculation engine applies classical Vedic rules as deterministic code, not as AI-interpreted guidelines. Yoga identification, for example, follows strict conditional logic: a Gajakesari Yoga is flagged only when Jupiter is in a Kendra (angular house: 1st, 4th, 7th, or 10th) from the Moon and neither planet is debilitated, combust, or in an enemy sign — per standard classical conditions. Similarly, Mangal Dosha evaluation checks Mars's placement from the Lagna, Moon, and Venus across the traditionally specified houses, with classical exceptions for sign and aspect conditions hardcoded in. These rules are not suggestions the AI interprets — they are boolean conditions the engine evaluates before the AI ever sees the data.

AI hallucination in astrology is not an edge case — it's the default behavior when a language model is asked to produce astrological content without factual grounding. A standard AI model, given a birth date and asked about career prospects, might generate a plausible-sounding paragraph about "Saturn's influence on your 10th house" without ever computing whether Saturn is actually in or aspecting the 10th house for that specific chart. The output sounds authoritative because the AI has been trained on millions of astrological texts, but the specific claims have no mathematical basis.

Keshoo's bounded architecture eliminates this category of error entirely. The AI synthesis layer receives a structured data envelope that contains every fact it is allowed to reference — and nothing else. When a user asks about career timing, the AI can see the 10th house lord, its placement, its Shadbala strength, the current Dasha-Bhukti rulers, and any transits activating career-relevant houses. It synthesizes a narrative from these specific data points. If a planet has no calculated connection to the career axis, the AI cannot manufacture one. The data envelope is the AI's entire reality for that reading.

This approach also prevents a subtler form of hallucination: interpretive drift. Even when an AI gets the planetary placement correct, it can generate interpretations that contradict classical Vedic principles — describing a debilitated planet as "powerful" or interpreting a combust planet as "highly visible," for instance. Keshoo addresses this by including contextual flags in the data package: debilitation status, combustion ranges, retrograde conditions, and friendship/enmity relationships between planets. These constraints guide the AI toward classically sound interpretation, not just factually accurate positioning.