AION Indian Market Intelligence

The problem

A single headline does not move the market in one direction. It moves different parts of it in opposite directions at once.

"Government raises import duty on edible oils" is bad for FMCG margins and good for domestic oilseed processors. Ask a conventional sentiment API and it returns something like negative, 0.82. Negative for whom? That number is not merely imprecise, it is the wrong shape of answer: it collapses a redistribution into a mood.

Impact has to land somewhere. Any tool that cannot say where is not analysing the event, it is reacting to the vocabulary in the sentence.

How it solves it

IMI takes one event — a headline, a circular, a notification — and returns where the impact actually lands, as structured JSON you can act on programmatically.

• The event, resolved to what actually happened rather than how it was phrased.
• Sector impact across NSE sectors as a signed vector: winners positive, losers negative, in the same response.
• Stakeholder views, because a producer, a consumer and a trader experience the same event from different sides of the trade.
• A confidence band and the supporting evidence, so you can tell a well-grounded answer from a weak one and decide how much weight it deserves.

It is built for Indian market structure specifically, rather than translated from a model trained on another country's economy. When the engine is unsure it says so instead of guessing, and when an event falls outside what it can resolve it declines rather than inventing an answer.

How to use it

You supply the news. We do not resell feeds. We do not store the text of your queries, except headlines the engine cannot classify: those are retained, linked to your API key by a one-way hash, and used to improve the engine, including by training its classifier. Usage is metered per account for billing using a one-way hash of each query. See the Privacy Policy.

Try it

Three free analyses per device on the public demo. Paste any Indian financial headline.

API

REST endpoint with a free tier: 10 requests per month, no card (verified email and passkey required). Paid plans from ₹999 a month. See APIs.

Agents

An MCP server is available so assistants can call the model directly.

Getting a good result

• Send the article body as well as the headline where you have it. Headlines are frequently ambiguous and the body is usually what resolves them.
• Treat the confidence band as part of the answer, not decoration.
• It interprets events. It does not tell you what to trade, and it returns no execution signals of any kind.