Intelligence & insights
The owner dashboard answers “is everything okay?” in one glance - collections, defaulters, live MIS. Every number is reproducible to the paise; forecasts and dropout-risk signals are deterministic and reason-coded (no black-box model); and any AI assist is grounded and human-gated - student data is never sent to a third-party model. Never a raw data dump.
See it live
Watch the real flow run
Know your school in two seconds - the AI shows its work.
The payoff
Outcomes, not features
Start to finish
Inside intelligence & insights
The real end-to-end flow - from the school's first action to the audited, reason-coded outcome.
- 1Roll the raw ledgers and rosters into KPIsThe owner dashboard sits on pre-aggregated snapshots, not a live scan of every table. Collections, defaulters and MIS come from deterministic roll-ups so the two-second read is fast and the same for every viewer, and money stays integer paise end-to-end, never a rounded float.
- 2Compute each number with a disclosed, reproducible methodEvery headline figure is emitted as a self-verifying artifact: the exact value, the method and formula that produced it, the labelled inputs it was computed from, the reason-coded factors behind it, and a sha256 reproducibility stamp. Anyone can recompute the stamp offline from the disclosed bytes and land bit-identical, so a number is bound to its stated reasons.
- 3Forecast and flag risk with explainable arithmetic, not a black boxForecasts use one of three named, disclosed methods (moving average, OLS linear trend, or seasonal-naive), each publishing its parameters and confidence band. Dropout risk is a transparent sum of named factors, and a minor is protected by default: the fee-arrears financial signal is excluded from a child unless a school explicitly opts in.
- 4Keep a human in the loop, and keep the model offA flagged risk becomes a trackable case with a frozen, reason-coded snapshot, and every lifecycle change requires a named human actor recorded in an append-only transition log; nothing is auto-actioned about a child. Any generative assist is grounded, human-gated and, by default, makes no third-party model call at all.
The building blocks
The modules that do the work
The 5 modules that make up intelligence & insights.
Reports & analytics
Owner dashboard, defaulter table, live MIS, CSV export - every number reproducible to the paise.
Explainable intelligence
Deterministic forecasts and dropout-risk signals, reason-coded - no black-box model.
AI assists
Grounded, human-gated drafting; no student data sent to a third-party model.
Analytics warehouse
OLAP roll-ups, KPIs and cross-campus benchmarking.
Field-proof dashboards
Read-only owner / auditor proof that the sealed ledgers recompute and reconcile.
Signature automations
What most systems don't do
Built-in mechanics you won't find elsewhere - running in the product, not drawn on a slide.
Three explainable forecast methods, no ML and no black box
Forecasting is deterministic statistics you can audit, not an opaque model. A forecast is a flat moving average, an ordinary-least-squares linear trend, or a seasonal-naive repeat, and every one discloses its method, window/season/horizon parameters and a widening confidence band. Same inputs always yield the same forecast, so an owner can reason about the number instead of trusting it.
Every number carries a reproducibility stamp
Each dashboard figure ships as an explanation object (value, method, inputs, reason-coded factors) sealed by a sha256 stamp over its canonical pre-image. Change one paise, one input label or one factor and the stamp diverges, and a verifier names the mismatch. It is reproducible AND tamper-evident: the number is provably bound to the reasons shown beside it.
Dropout risk is a transparent sum, with fee-arrears excluded for minors
Risk is not a mysterious score; it is the sum of named, fired factors you can read line by line. Because a child fee-arrears signal is a financial attribute under DPDP Section 9, it is excluded from a minor by default (weight zero, and the child fee state is not even read), included only on an explicit, fail-closed per-tenant opt-in.
Zero third-party LLM egress by default, enforced by a PII guard and a fetch-trap test
The AI-assist runtime makes no third-party model call on its default path, and a guard refuses any prompt object carrying a name-like or contact key (keys only, never values). A DB-backed test installs a fetch trap that fails if any byte tries to leave. The one optional external assist (report-studio template design) is default-off and only ever receives opaque field NAMES, never a data value.
Also built in
More in intelligence & insights
Capabilities that already ship in the product - each shown live, not just named.
Integration & developer API
Signed API and webhooks that keep you in sync
Vidyom exposes a versioned REST API and outbound webhooks so your group ERP, government portals and BI tools stay in sync automatically. Every request and every event is HMAC-signed end to end - no PII leaves your control.
- Versioned REST endpoints with predictable JSON contracts.
- Webhooks fan out each event to every subscribed system.
- HMAC signature on every request and delivery, verified fail-closed.
Straight answers
Intelligence & insights - common questions
Is this a black-box AI that just gives me a score?
No. The intelligence is deterministic and reason-coded. Forecasts use three named statistical methods, and every number carries its method, inputs and a reproducibility stamp you can recompute yourself. There is no ML model deciding anything you cannot inspect.
Does the dropout-risk signal punish poor families?
Not for a minor by default. The fee-arrears financial factor is excluded from a child unless a school explicitly opts in, and when excluded the child fee state is not even read. The score is a visible sum of named factors, and any resulting intervention is opened and acted on only by a named human.
Does any AI feature send my students to an external model?
By default, no external model call happens at all, and a PII guard plus a fetch-trap test enforce it. The single optional external assist, report-studio template design, is off unless a school turns it on, and even then it receives opaque field names, never student values.