Skip to content
How it works

From seven messy exports to one clear next step

Each stage is measured, explained and visible inside the product. Nothing is a black box.

  1. 01

    Ingest

    CSV, Excel and JSON exports from ERP, attendance, LMS, clubs, placement, skills and feedback. Headers are mapped automatically.

  2. 02

    Match

    Roll numbers in five formats are normalised. Names and emails are fuzzy-matched; uncertain links wait for a person.

  3. 03

    Score

    Every indicator becomes a cohort percentile, then six weighted components make the 0 to 100 Success Score.

  4. 04

    Predict

    Two calibrated models estimate academic and placement risk separately, with SHAP drivers for each student.

  5. 05

    Act

    Mentors get a ranked weekly plan and a suggested intervention. Every action is logged with a baseline.

  6. 06

    Audit

    Accuracy, calibration, stability and fairness are measured and shown inside the product.

The seven exports

What goes in, and what is messy about each

Seven files from seven campus systems, no shared ID and no shared format. This is the mess Ascora joins into one record per student.

  1. 01CSV

    ERP academics

    Marks, internals, attendance and backlogs per subject and semester.

    The mess: 180 duplicate rows, 1,158 missing internal marks, 12 out-of-range values, and 12 students listed under two branches.

  2. 02Excel, 3 monthly sheets

    Attendance register

    Classes held and attended per subject, per month.

    The mess: Merged header rows and 144 student-months with no entry.

  3. 03CSV

    LMS activity log

    Weekly logins, minutes online and assignment completion.

    The mess: 144 email variants, 24 students with two accounts, 2 date formats and 557 missing minutes.

  4. 04CSV

    Clubs and events

    Hackathons, workshops, certifications and club roles.

    The mess: Identified by name only, with 3 date formats and 622 initials or spelling variants.

  5. 05Excel

    Placement cell

    Aptitude and coding scores, mock interviews and placement-cell visits.

    The mess: 36 duplicate rows, 6 out-of-range scores, and 305 students who never visited (a gap, not a failure).

  6. 06Nested JSON

    Skills assessment

    Technical and soft-skill scores per student.

    The mess: Nested records, and 151 of 1,139 scored on a 0 to 10 scale instead of 0 to 100.

  7. 07CSV

    Feedback forms

    Course ratings and free-text comments.

    The mess: 424 anonymous rows and comments in English, Hindi and Hinglish.

These counts describe the generated sample cohort of 1,200 students, built to look like real campus exports. A real college would bring its own files.

The Success Score

A transparent formula you can read in one line

Six components, fixed weights, all in one config file so a college can tune them.

  • Academic 30%

    CGPA, internals, trend vs own past, backlog penalty

  • Placement readiness 20%

    Aptitude, coding, mock interviews

  • Attendance 15%

    Overall and core-subject attendance

  • LMS activity 15%

    Logins, assignment completion, login trend

  • Skills 10%

    Technical and soft-skill assessments

  • Engagement 10%

    Hackathons, certifications, workshops, clubs

Success Score = 0.30 Academic + 0.20 Placement + 0.15 Attendance + 0.15 LMS + 0.10 Skills + 0.10 Engagement

Models

Measured on a held-out test set

A plain logistic baseline scores close to the boosted model. We say so; the gain is calibration and per-student explanations.

Academic risk AUC
0.881
logistic baseline 0.875
Placement risk AUC
0.802
logistic baseline 0.806
Precision at top 50
0.78
academic model
Rank stability
0.997
worst case over 12 weight shifts

All figures come from the generated sample cohort of 1,200 students. Real use needs retraining on a college's own outcomes.

Design choices

Decisions we made on purpose

Feedback never lowers a score

An unhappy student must not be penalised for saying so. Feedback is shown as context only.

Missing data is a gap, not a failure

No placement-cell visits means a neutral-low readiness and a clear 'no data' label, not a zero.

Two risks, never one label

Academic and placement risk are separate. Students aiming for higher studies are not flagged for placement.

Sensitive traits are never inputs

Gender, residence, first-generation status and similar fields are used only to audit fairness.

From early warning to real intervention

Sign in as a head of department or a faculty mentor and walk a full cohort of 1,200 students.

Open AscoraHow it works
How it works | Ascora