Methodology

Activities published

26,090

Composition

19,088 publications; 2,400 invited lectures; 2,078 faculty development programmes; 729 patents; 697 memoranda of understanding; 367 events organised; 284 professional bodies; 247 funded projects; 141 awards; 59 consultancies

Activity dates

2000 to 2027

Figures generated

9 November 2026

Activity dates are when each activity took place, not the year it was reported.

1. 1CollectScopus, PBAS data and the MoU register
2. 2MergeDuplicate reports become one record
3. 3ClassifyAgainst the SDGs and Viksit Bharat themes
4. 4ReviewChecked, then published by IQAC
5. 5PublishShown on this site

Scopus

Elsevier’s research database. It supplies the university’s published research, including KIIT and KIMS: title, authors, year, journal, keywords and funder.

PBAS data

The Performance Based Appraisal System returns filed by each school. They cover events, faculty development programmes, invited lectures, patents, funded projects, consultancies, awards and memberships.

MoU register

The Internal Quality Assurance Cell’s register of agreements with partner institutions, giving the partner, its country and the term of the agreement.

Every record names its source. Evidence is a link to the original, such as a journal or a patent registry.

The same activity is often reported more than once: in different years, by several faculty members, or twice in one sheet. These copies are merged into a single record.

Patents are matched on application number, projects on sanction reference and events on name and date. Where a match is uncertain, the records are kept apart rather than merged.

Each activity is classified against two frameworks: the 17 UN Sustainable Development Goals with their 169 targets, and the 5 Viksit Bharat @2047 themes with the sub-themes beneath them. Classification follows the written rubric below, and every label names a single node in one of those frameworks: a goal, one of its targets, a theme or a sub-theme.

1. 1ComposeThe record’s fields, in a fixed order
2. 2JudgeAgainst the rubric, by the instrument for its type
3. 3LabelNode, primary, rationale, confidence, versions
4. 4ThresholdModerate or above is published

What the classifier reads

The classifier reads a composed record rather than a title. For an activity the input carries its title, type, level, school, location and, where the record holds them, its outcome, its host and its collaborating agency. For a publication it carries the abstract, the author keywords and the funding body. The fields always arrive in the same order, fixed by a versioned input template, so two runs under the same template saw the same shape of input.

Instrument by record type

Different kinds of record carry different amounts of usable evidence, so each is classified by the instrument that suits it. All of them work to the same rubric.

What every label records

Node

The goal, target, theme or sub-theme. A target is preferred over a bare goal.

Primary

Exactly one label per framework is marked primary: the goal the activity is actually about. The others are secondary.

Rationale

One sentence naming the words or fields that decided the label. A rationale must be at least 60 characters, must not be a generic restatement of the label, and may not be reused word for word across unrelated records. One that fails any of these checks is refused before it is written.

Confidence

One of four written levels, set out under the rubric below.

Versions

The model, the input template, the rubric, the taxonomy and, for a rule, the lexicon it came from. A rubric revision applies to the work that follows it; earlier labels keep the version they were made under.

Each activity page shows all of this for every label it carries.

Viksit Bharat labels

A Viksit Bharat label is reached in one of two ways. Most are derived from the activity’s primary SDG through a fixed crosswalk that links each goal to the themes it serves. Where a sub-theme is clear but no goal leads to it, such as women’s empowerment, culture and the arts, cybersecurity or sport, the sub-theme is judged directly. Each derived label says so in its own rationale.

Where each label came from

Rubric version 1.1.0

The rubric is the written instruction every classifier works to. It is held as code, versioned, and stamped on every classification run. Any change to it produces a new version.

The system prompt

This is the complete system prompt given to the AI model that classifies each publication. The first part is the rubric’s instruction; the rest adapts it to a publication and fixes the format of the answer. The user message that follows it is the composed record described under Classification, one publication at a time. With each batch the model also receives the written definition of every confidence level, summarised below.

You are classifying a single record of university activity against two frameworks:
the 17 UN Sustainable Development Goals (with their 169 targets) and the five
Viksit Bharat @2047 themes.

Rules:

1. Judge only what the record says. Do not infer institutional intent, do not
   credit an activity with an outcome it does not claim, and do not reward a
   record for the university's reputation in a field.

2. Multi-label is expected and correct. An activity may legitimately support
   several goals. Mark exactly one as primary per framework: the goal the
   activity is actually about, not the most impressive one it touches.

3. Prefer a target over a bare goal wherever the record supports it. "SDG 4"
   is weaker evidence than "SDG 4.3", and the target is what a reviewer can
   check.

4. Assign a confidence band from the four defined bands. Do not invent a
   numeric score.

5. Write one sentence of rationale naming the specific words or fields that
   decided it. "Mentions renewable energy and smart grids" is useful.
   "Relates to sustainability" is not.

6. Where the record cannot be classified - a title of "FDP" or "AI" with no
   other usable field - return no labels and say so. Roughly 144 records in
   this corpus have titles of ten characters or fewer. A gap is honest and
   correctable; a guess is neither.

7. Do not assign SDG 17 from text. Partnership is detected structurally by a
   separate deterministic rule, so that rule-derived and judged labels stay
   separable in the methodology. If a record's subject genuinely is
   international cooperation as a topic, say so in the rationale and let the
   rule decide the label.

You are classifying a peer-reviewed publication indexed in Scopus, not a PBAS activity record, so "the record" below means the paper: its title, journal, subject keywords and abstract where present. Many records here carry a full abstract; where one is present, weigh it over the title. Where it is absent, judge honestly on what is given rather than guessing from the journal name alone.

Respond with nothing but a single JSON object, no markdown fences, no
commentary before or after it, matching exactly this shape:

{
  "sdg": [ { "node": "4.3", "band": "high", "rationale": "one sentence" } ],
  "vb": [ { "node": "vb-1.2", "band": "high", "rationale": "one sentence" } ],
  "unclassifiedReason": "one sentence, present only when sdg is []"
}

"node" is an SDG target code ("4.3") or bare goal code ("4") if no target fits, or a Viksit
Bharat theme/sub-theme code ("vb-1", "vb-1.2") for the vb array. "band" is exactly one of
"certain", "high", "moderate", "low". "sdg" holds at most three entries, the first
being primary. Never put SDG 17 in "sdg" - rule 7 explains why. "vb" holds at most one entry;
omit it entirely (or leave it []) when no Viksit Bharat theme is a strong direct fit - do not
force one. Every rationale names the specific words or fields that decided it, per rule 5.

Confidence levels

Confidence is one of four written levels rather than a numeric score. A figure such as 0.83 looks like a measurement but encodes no real distinction, while four levels with written criteria can be applied consistently and checked by a second reader. Labels at Moderate or above are published.

1. LowThe evidence is too thin to assert the goal. No label is written, and the activity is marked unclassified against that framework.Not published
2. ModeratePlausible and defensible, but resting on an assumption a reasonable reviewer could decline.Published
3. HighA strong, direct inference from the record type, the funder or the programme, even where the title is thin. An ICMR-funded project against SDG 3.Published
4. CertainThe record states the subject explicitly. A workshop titled “Renewable Energy and Smart Grid” against SDG 7.Published

Nothing reaches this site automatically. Every record starts as a draft, and only an authorised member of the Internal Quality Assurance Cell can publish it. Each publication is logged, and a record can be withdrawn the same way.

Before a record is published, the system checks that it is classified, that it is not waiting on an unresolved merge, and that it traces back to its source. A record that fails a check is held back.

The rubric itself is written and maintained by the Internal Quality Assurance Cell.

Activities are classified against the 5 development themes of Viksit Bharat @2047:

• Empowered Indians
• Thriving and Sustainable Economy
• Innovation, Science and Technology
• Good Governance and Security
• India in the World