सिया
Siya
A pilgrim helpline built on 124 sourced knowledge-base entries across 32 categories, with live official-source lookups and an explicit refusal to answer what nobody has published.
- Languages
- Hindi · English
- Channels
- Inbound phone · Browser voice call
The problem
The information a devotee needs about Ayodhya is published, contradicted and republished. Darshan timings were revised to a seven-slot schedule while older pages still carried the previous hours. Aggregators and guide sites publish tariffs and timings nobody has verified. Someone planning the trip from another state cannot tell which page is current.
That ambiguity has a cost beyond inconvenience. Every darshan and aarti pass is free on the official portal, and the temple trust has publicly warned that anyone charging for one is committing fraud — but the warning only reaches devotees who already found the official site.
And the people with the most questions are the least served by a website. Elderly devotees, first-time travellers and families arriving from villages are asked to navigate booking portals, station enquiries and helpline menus. Reading is the barrier. Speaking is not.
What AiSewak built
सिया answers pilgrim questions by voice from a knowledge base of 124 entries across 32 categories of question, built only from primary sources: the temple trust, the Ayodhya district administration, UP Police, the India Meteorological Department, Indian Railways, UPSRTC, UP Tourism and PIB. Each entry carries its source and its verification date.
Anything newer than the knowledge base is fetched during the call — today's timing changes, a VIP-visit closure, slot availability, festival dates, weather and flood warnings, train status — from official sources only, never from aggregators.
The design decision that matters most is the third one. The knowledge base records what is genuinely unpublished — parking tariffs, the Saryu aarti timing, a standalone dress-code order — and on those the agent says the official information is not available rather than producing a plausible answer. A confident wrong timing given to someone already travelling is worse than an admission.
Coverage spans darshan and aarti timings, the free Sugam Darshan pass and what identification to carry, entry and security rules, prasad and the official donation channels, air, rail and road travel, local transport, stay and community kitchens, accessibility for elderly and Divyangjan devotees, medical posts, emergency numbers, lost property and missing persons, and seasonal and festival arrangements.
How it runs
Conversational layer
An ElevenLabs Conversational AI agent, Hindi-first with a language lock, provisioned from a checked-in build script so the agent configuration is reproducible rather than hand-edited in a console.
Live official-source lookup
A dedicated topic on the shared search endpoint, with its own allowlist of official Ayodhya and Uttar Pradesh sources, so a pilgrim question never resolves against a travel aggregator.
Non-affiliation, stated by the agent
The agent declares that it is an AI assistant and that it is neither the Shri Ram Janmbhoomi Teerth Kshetra Trust nor a government helpline. It never claims to be a priest, an official or a human, and the page carries the same disclosure above the demo.
No money, and no fake help
It books nothing, issues no pass and accepts no payment or donation; it will not advise tipping or paying anyone, and it repeats the trust's warning that paid or VIP darshan does not exist. A medical emergency, accident, crime or missing person routes to 112 before anything else, with no lookup first. Every call is scored for compliance automatically.
What a department can replicate
- 1
Any temple trust, district administration or tourism department running a visitor line can lift the sourcing discipline directly: build the knowledge base only from primary sources, store the source and verification date alongside every entry, and treat an unverified aggregator figure as absent rather than as a fallback.
- 2
The second reusable piece is the explicit unknowns list. Enumerating what is genuinely unpublished, and wiring the agent to say so, converts the most dangerous failure mode of a language model into a scripted, auditable answer.
- 3
The third is the emergency short-circuit: a defined set of situations where the agent stops answering questions and gives a verified number, before anything else in the conversation runs.
Scope the same pilot for your department
Two districts, a bounded knowledge base, a written list of what the agent will refuse to do, and a measurement step in the call itself. That is the shape every pilot above started from.