Every system you run, in one governed foundation.

Every system your institution runs flows into one governed foundation.
Intelligence, agents and people work from it, on the lake and providers you choose.

From source record to approved action, in seven steps.

Every record follows the same governed path through Vedhra.
Each step shows what happens, what it guarantees, and what you see in the product.

1Connect

Sync every source into your lake.

Each source syncs into your own data lake on a schedule, copying only records that are new or changed since the last run.
Records land unchanged and append-only, so any run can be replayed, and Vedhra never queries your live systems.

Under the hoodAPI and webhookssource contract v3append-only raw landingreplayable

ingestion · all sources · last 24 h14 sources synced

14sources

4.8Mrecords landed

0failed runs

Sources

SalesforceCRM · change data capture412k

MoodleLMS · API + webhooks2.1M

SAPFinance · nightly batch186k

SISStudent Information SystemAPI · every 15 min94k

TPTally PrimeAccounts · on-premise agent38k

XLSpreadsheets and data dumpsSFTP · CSV, Excel12 files

DODocumentsExtraction · OCR1,840 files

Raw landing · append-only

s3://vedhra-raw/
├ crm/ · sis/ · lms/ · finance/ · hr/
├ files/ · documents/
└ {source}/{entity}/dt={date}/part-*.parquet

2Refine

Refine through tested stages.

Records move from Raw to Bronze (typed), Silver (cleaned and deduplicated), Canonical and Gold (business-ready).
Every stage runs schema, uniqueness and freshness tests. Failing rows are quarantined, never promoted.

Under the hoodmedallion stagesschema and uniqueness testsquarantine on failurefreshness SLA

pipeline · nightly run 4412 · all sourcespassed · 6 min 12 s

Raw✓ counts match every source4.8M records

Bronze✓ typed to source contracts214 tables

Silver✓ cleaned · 61k duplicates removed186 tables

Canonical✓ every row linked to a vedhra_id42 entities

Gold✓ fresh · within SLA63 tables · 118 metrics

!1,204 rows quarantined: 3 source contract violations. Not promoted.

3Resolve

Resolve one identity per person.

Rules and probabilistic matching link records for the same person across systems, each with a confidence score.
Matches below threshold go to a reviewer queue, and every merge is recorded and reversible.

Under the hoodrule-based and probabilistic matchingconfidence scoresreviewer queuereversible merges

identity · 31,482 people resolved5 of 5 linked

97.8%matched automatically

214in reviewer queue

0.94lowest auto-match

Example · five records, one person

LMSarao24email · exact

AdmissionsAnanya Raoemail · exact

Student Information System24BBA0187enrolment no. · exact

FinanceRao A. - BBA 2024name + programme · 0.94

HRA. Rao · TAreviewer confirmed

vd_8f41c2Ananya Rao · BBA · 2024 intake

4Model

Map into one canonical model.

Every source maps once into an education model of people, programmes, courses, terms, enrolments, fees and outcomes.
Metrics are defined once, versioned and owned, so every report and agent computes them the same way.

Under the hoodcanonical education modelversioned metric definitionsinstitution extensions

canonical model · v342 entities

Core entities

person31,482

student11,874

staff1,206

programme48

course · section1,312

enrolment68,210

assessment412k

fee_position11,874

placement2,340

metric.active_studentv3 · owner: RegistrarRegistered for the current term, with fees cleared or on an approved plan.

5Trace

Trace every value to its source.

Column-level lineage links each Gold value to the rows and transformations that produced it.
Open table formats keep every version, so any figure can be reproduced as of a past date.

Under the hoodcolumn-level lineagepoint-in-time queriestime travel on open tables

lineage · active_students = 11,874traced

today · 11,874as of 30 Sep 2025 · 11,612

Goldgold.metric.active_student

11,874 · current term

Canonicalenrolment ⋈ fee_position

definition v3

Silversis.enrolment · finance.ledger

12,106 registered · 232 fees pending

Bronzesis.student · sap.fi_document

typed as landed

Raws3://vedhra-raw/sis/… · /finance/…

2 sources · 14 files

6Reason

Detect, explain and recommend.

Models and governed retrieval run on the canonical layer, within each user's permissions.
Every output carries its evidence, a confidence score and the data that would improve it.

Under the hoodgoverned retrievalpattern modelsevidence and confidencedata readiness

signals · this week3 new

Attendance pattern, BBA year 214 students below 70% for three weeks. Advisor meetings recommended.conf 0.86

Hostel demand, BSc Data ScienceYear 3 demand exceeds capacity by 140 beds.conf 0.91

Fee plans, Semester 3Missed instalments up from 3% to 6%. Earlier reminders recommended.conf 0.78

Every signal includesEvidence and sources, what changed, a recommendation with confidence, and who acts next.

+Connecting counselling notes would raise confidence on the attendance signal to about 0.93.

7Act

Act through approval and write-back.

Agents propose changes, and a named approver signs off by role and threshold.
The change is written through the owning system's API, read back to confirm, and logged end to end.

Under the hoodapproval by rolewrite-back through the owning APIread-back confirmationfull audit trail

proposal pr_0412completed

Schedule advisor meetings for 14 students

✓ProposedEarly Alert Triage agent · with evidence

✓ApprovedDr Meera Iyer, Programme Head · 10:12

✓WrittenAdmissions · 14 meetings created

✓Confirmedread back and reconciled · rcpt_7c2e19

WhatsApp · to studentHi, you have an advisor meeting on Thursday at 11 am with Dr Meera Iyer.

The detail behind each step.

More than 120 connectors are built on six patterns. Each reads to a source contract and lands data unchanged.

PatternUsed forFreshnessExamples

API connectorsREST · GraphQL · OData Modern SaaS and cloud platforms every 15 min SalesforceMicrosoft DynamicsZohoMoodleCanvasDDarwinboxRazorpay

Change data capturelog-based CDC · platform events High-volume operational databases and CRMs every 5–15 min Salesforce CDCPostgresMySQLSQL Server

Webhooks and eventspush · signed payloads Payments, forms and event streams every 5–15 min Payment gatewaysapplication formsLMS activity

Batch filesSFTP · CSV · Excel · XML Exports, spreadsheets and data dumps scheduled Exam resultsscholarship listslegacy reports

On-premise agentoutbound-only TLS 443 Software with no modern API, inside your network scheduled TPTally Primeon-premise ERP and exam systems

Document extractionOCR · layout · schema mapping Unstructured files and scanned records on arrival Transcriptsoffer lettersaccreditation evidence

Your data stays in formats you own.

Tables sit in open formats, in storage you choose, registered in a catalog you control.
You can change cloud, query engine or model provider without starting again.

Vedhra tells you what data each goal needs.

Choose an outcome your institution cares about.
Vedhra shows what it can already see, what is missing, and what to connect next.

Reach it on any channel. Agents act with approval.

Students, parents, staff and leaders reach the same foundation through the channels they already use.
Agents do the work, and every consequential action has a named approver.

Interfaces and communications

WhatsApp · Ananya Rao

KAIHi Ananya, your attendance in Business Statistics is 68% this month. Would you like to meet your advisor this week?

AnanyaYes, Thursday works.

KAIBooked: Thursday, 11 am with Dr Meera Iyer. I have added it to your calendar.

VoiceWeb chatWhatsAppTeamsEmailSMS

Also leadership briefs alerts dashboards MCP · SQL · API

Agentsat work

42agents across nine families, from enquiry to alumni

AttractEnquiry Concierge118 calls booked

ThriveEarly Alert Triage14 students with a plan

StewardLeadership Briefing Agent4 decisions briefed

✓Each agent proposes, a named person approves, and the action is written back to the system that owns the record.

View all agents →

Start with your own systems.

Bring two or three systems and one question.
We will show you Vedhra working on your data.