AI & Ethics
● Demo dataExplainability, governance, and human-in-the-loop oversight for every AI-assisted output. AI explains, synthesizes, and recommends — people decide.
AI Decisions Logged
0
Pending Human Review
0
High-Stakes Flagged
0
Avg Confidence
0%
Source-Cited
0%
AI Guardrails (enforced)
Do not fabricate data — answer only from the institution’s integrated dataset.
Do not answer beyond available data; refuse when there is no grounded basis.
Every answer cites its internal data sources.
Every answer shows a confidence score.
Every answer states known limitations.
Every high-stakes recommendation requires human review before action.
AI explains, synthesizes, and recommends — it does not autonomously decide.
Explainability — how scores are derived
Final Match Score
0.30 Research + 0.20 Faculty + 0.15 Intl-Collab + 0.15 South-Asia Fit + 0.10 Funding + 0.05 Mobility + 0.05 Network proximityFaculty Alignment
0.30 citations + 0.25 h-index + 0.20 publications + 0.15 intl-collab + 0.10 recencyEvery Ask DRISHTI answer
returns its data sources, a confidence score, and known limitations — and refuses when no grounded basis exists.Decision & Recommendation Log
Every AI-assisted output is traceable — with model, confidence, cited sources, and human-review status. High-stakes recommendations are not auto-actioned.
0 records
| When | Type | AI output | Model | Confidence | Sources | Human review | Reviewer |
|---|---|---|---|---|---|---|---|
| No records match your search. | |||||||
Bias & Coverage Monitoring
Average data confidence by region — surfaces where the model is least certain so recommendations there get extra scrutiny.
Coverage Gaps — low-confidence markets
No markets below the 75% confidence threshold.
Model Governance Registry
PRAGATI weighted scoring
Purpose: Match & prioritization scores
Inputs: Research, faculty, mobility, funding, South-Asia fit, network proximity
Basis: Deterministic weighted model
Transparent formula · human-reviewed
Ask DRISHTI (grounded retrieval)
Purpose: Executive question answering
Inputs: Curated Q&A over golden-record data
Basis: Retrieval, not generation of facts
Cites source + confidence + limits; refuses if ungrounded
Entity resolution
Purpose: Deduplicate records into golden copies
Inputs: Name/attribute similarity across sources
Basis: Rule-based + AI-assisted matching
High-similarity auto-merge; ambiguous held for review
Risk & gap classifiers
Purpose: Flag stewardship gaps, dormant partners, low confidence
Inputs: Activity, confidence, recency signals
Basis: Rule-based thresholds
Advisory flags; human action required