About Prajna न्याय

Prajna (Sanskrit: न्याय, “justice”) exists to make the Indian justice system legible and accountable. We aggregate publicly available information about cases of violence against women, normalise it into a structured timeline, and surface systemic delays so journalists, researchers, and civil society organisations can act on them.

We never publish victim identities. We are not affiliated with any government body. Data is derived solely from public sources.

• ANI / PTI: newswire RSS feeds, English
• Dainik Bhaskar: Hindi regional press
• Mathrubhumi: Malayalam press
• Ananda Bazar Patrika: Bengali press
• eCourts API: official court order data
• NCRB: National Crime Records Bureau annual Excel/PDF reports
• RTI responses: scanned PDF uploads processed via OCR

All sources are attributed per event. Source confidence scores are displayed on each timeline entry.

Raw articles pass through a four-stage pipeline before appearing on this platform:

1. Privacy Engine: Detects and redacts names, ages under 18, phone numbers, email addresses, and precise addresses. Minor detection triggers full suppression per POCSO guidelines.
2. AI Extractor: A large language model (claude-sonnet-4-6) extracts structured events at temperature 0. Every event requires a verbatim “source quote” from the original text to prevent hallucination.
3. Entity Resolver: Links new events to existing cases using FIR number matching, court record matching, and multilingual sentence-embedding similarity. Ambiguous matches go to human review.
4. Timeline Engine: Places each event into one of seven legal stages (FIR → Investigation → Chargesheet → Trial → Judgment → Appeal → Closure) and flags statutory deadline breaches.

Every event carries a confidence score (0–1) based on source trust, field completeness, and model certainty. Events below 0.60 are archived and not shown publicly. Events between 0.60–0.90 are queued for human moderator review. Events above 0.90 from trusted sources are auto-approved.

Cases with an overall confidence below 0.85 are marked “Under review” in the UI.

• Victim names are replaced with HMAC-SHA256 pseudonyms (e.g. VICTIM-3a9f2c).
• Any article triggering minor-detection (POCSO / age < 18) is fully suppressed from public view.
• Precise addresses are masked; only state and district are retained.
• Phone numbers, email addresses, and national ID numbers are redacted before storage.
• Under India’s DPDP Act 2023, erasure requests may be submitted to privacy@nyaya.org.in.

Prajna flags delays against the following legal deadlines:

Prajna is licensed under the GNU Affero General Public License v3.0. Source code is available on GitHub.

Contributions, bug reports, and data corrections are welcome. Researchers requiring bulk data exports may contact us for API key access.