Teardown: EduNavigator AI

Two hosts cover EduNavigator AI, a demo system built against a fictional Ludhiana study-abroad consultancy, and why the hybrid answer — explanation plus a live-queried table — is the actual product.

A: Same disclosure as always for these — this is the company's own build, against a fictional study-abroad consultancy in Ludhiana. Not a client deployment.

B: The premise: students ask questions that need both narrative and data at once. "What fits my budget for a January intake?" And generic chatbots fail in one of two directions — fluent prose with an invented deadline, or an accurate spreadsheet nobody reads.

A: Both fail exactly where the cost is highest — admissions decisions, not casual browsing.

B: The core decision in the build: one response streams a grounded explanation with numbered citations, then a table captioned to say plainly what it is — matched records, queried live against the current catalog, not generated by the model.

A: The obvious approach would've been letting the model describe the matching programs in prose, the way a general assistant does by default.

B: Rejected, because prose is exactly where invented deadlines and invented tuition figures sneak in. A model asked to describe a table in words is a model asked to paraphrase numbers — and paraphrased numbers are where the risk concentrates. Deriving the table at request time removes that step entirely.

A: Second decision — one dataset, two apps. The student-facing explorer, its chat answers, and the consultancy's admin catalog manager all read the same shared data module. Editing a record in admin is editing what the assistant says five minutes later.

B: The obvious approach would've been a separate content pipeline — counsellors update a spreadsheet, someone periodically re-ingests it. Rejected, because a consultancy's credibility depends on freshness. A tuition figure verified last month may already be wrong, and a re-ingestion step is a delay between a correction and it actually taking effect.

A: One dataset means that delay is zero.

B: Third — market-specific authenticity, not just translation. Tuition's always shown in the actual budget figure students think in. The vocabulary students and counsellors already use, not a generic glossary. The admin console runs on the local timezone.

A: The obvious approach would've been building a generic version and treating the local market as a configuration flag. Rejected — a consultancy's clients can tell within one exchange whether the system understands their situation.

B: Getting the vocabulary generically correct but locally wrong reads as exactly the kind of tool that doesn't understand the market it's deployed into.

A: What's the useful takeaway, if your business isn't study-abroad advice?

B: If you give advice that needs an explanation and current data in the same breath — pricing, availability, eligibility — and today those two things live in different places that can drift apart, this pattern's worth evaluating. One dataset behind both.

A: One thing to check this week.

B: Find one place in your business where an explanation and a number are supposed to agree. Ask when each was last updated, separately. If the answer's two different dates, that's the gap.

A: Written version's at atinnovators.in, under insights.