Why we built TathaAstu
Hindu calendar calculations sit underneath a great deal of software. Almost none of it can explain the dates it displays.
The problem
Hindu calendar calculations sit underneath temple apps, matrimony platforms, festival calendars, payroll, travel, publishing and now AI assistants. Yet the APIs serving them return a date and nothing else. When the date is disputed — and it is disputed often, because these are questions communities care about — there is nothing to inspect.
What we changed
We rebuilt the festival engine around shastric kala-vyapini rules operating on real tithi intervals, rather than matching whatever tithi happened to be running at sunrise. Then we made the engine show its work: every festival carries the rule, the version, the authority, the overlap in minutes, and a confidence score.
Why it matters
A date you can audit is a date you can defend. An HR department can answer an employee. A temple committee can review a schedule. A newsroom can cite a source. An AI agent can ground an answer instead of inventing one.
What we will not do
We will not publish a number we have not measured, and we will not hide a limitation to look better. The Accuracy page lists our known problems because you are going to find them eventually, and it is better that you find them from us.
History
How the platform got here
Eight milestones, in order. The interesting one is M4 — the point at which festivals stopped being a lookup table.
Foundation
Initial API surface, panchang data import across 14 cities.
Auth and usage
JWT auth, API key management, plan-based rate limiting, usage tracking.
Astrology
Birth chart, Guna Milan compatibility, PDF reports, matchmaking engine.
Festival Engine V2 — core
KalaCalculator, nine evaluators, TieBreakEngine, rule versioning, schema migration.
Festival Engine V2 — cutover
Tier A/B/C validation, shadow mode, production cutover.
Platform alignment
Tier enforcement, webhook platform, plans and billing, analytics.
Security hardening
Credentials moved to environment, secrets rotated, 152 tests at 91% coverage.
Verification framework
Execution engine, comparison policies, adapter registry, external benchmark.
A note on how we talk about accuracy