YOJNA 1.0
For everyday answers.
- Reads your situation and what is urgent in it.
- Finds schemes and jobs it can trace to a page, and says when one is not yet verified.
- Explains the plan computed from your facts, and what changed when they change.
YOJNA 1.0 and YOJNA 1.0 Pro.
YOJNA’s own in-house AI models. They are not for the public — not sold, not licensed, no API. Like every language model they predict the next word; what sits around them, below, is what makes an answer YOJNA’s.
What each model does.
YOJNA 1.0 Pro
For the harder questions.
- Money and legal maths, comparisons, questions with several parts.
- Slower — it thinks harder before it answers.
- The deeper answers in the higher plans come from here.
What is next
Next: models we train ourselves on the work of understanding real situations. That is where this is going, not where it is today.
What is ours.
- Verified catalogue
- Schemes and jobs, re-checked over time. Each links to its official page where one exists, and where a link has been checked it shows the date.
- Memory
- What you told it, kept apart from what it worked out. Visible and correctable, and it forgets on request.
- Eligibility rules
- Decided by rules, never by a model. When a detail about you is missing it asks — a gap is never a no.
- Evaluation
- Regression suites and side-by-side runs, on the same path a person uses.
- Safety rules
- What the models may not do, and where a person must confirm.
The architecture
The model is not the product. YOJNA is the product.
A language model writes the sentences. It does not decide what is true about you, what you qualify for, or what happens next. Nine layers around it do — and those are ours.
- Situation
- What you type in your own words is resolved to the situation you are actually in.
- Memory
- What you told it is recalled and kept apart from what it worked out, each with its source.
- Routing
- Every request is sent to the right class of reasoning — YOJNA 1.0, or 1.0 Pro for the hard ones.
- Tools
- A planner picks tools that read YOJNA’s own data instead of letting the model guess.
- Eligibility
- Rules in the database decide what you qualify for. A model never issues that verdict.
- Verification
- Listings are checked against their official pages, and labelled when a check is still pending.
- Plan
- Your facts become steps with dates, and the plan is recomputed when a fact changes.
- Safety
- A screening layer on what comes in, and nothing done in your name until you confirm.
- Evaluation
- Changes are measured on the work itself, and what actually happened to you feeds back.
Swap the language model and YOJNA still knows you, still checks the page, still holds the rules. That layer is a component. The system around it is the company.
How an answer is made.
Your situation
What you said, what it already knows about you, and what is urgent.
Options, with sources
Each one shows where it came from, or says plainly that it is not yet verified.
A plan
Today, this week, this month, in order.
Updates
When a fact changes, the plan is recomputed. Every action in your name is confirmed by you.
Both models work this way. The deeper answers come from YOJNA 1.0 Pro.
Safety and limits.
- You press submit
- Nothing is sent in your name — YOJNA stops at the submit button and waits for you. And never pay anyone to press it for you.
- Unknown is not no
- When a detail is missing it asks; when something is uncertain it says so. A gap is never a rejection.
- Official pages win
- Where an official page exists it is linked. If the two disagree, the official page is right.
- Your information
- Encrypted in transit, never sold. See it, correct it or forget it from inside the app.
How it is evaluated.
We do not publish benchmark numbers we have not run.
What we check
- Did it collect the facts before it advised?
- Did it stay inside its rules?
- Did the plan follow from the facts?
You don’t need to have everything figured out.
Tell YOJNA what is happening. It remembers, checks, and comes back.