Aegis is an AI school that does research the way engineers do: by shipping systems with public institutions and measuring what happens. Four live programmes below; each one is also where our students learn to build AI responsibly.
AI ranks, suggests and flags. An officer, a reviewer or a teacher makes every consequential decision, and the system records who.
Every decision carries a signed, reviewable evidence packet: inputs, thresholds, model version, reviewer action.
Public data is published as open JSON for researchers; personal data is never searchable by the public.
India-hosted or on-prem, tenant-scoped storage, and interfaces in Indian languages that work without a signal.
“Watching Maharashtra's rivers, creeks and coast.”
Maharashtra has 54 polluted river stretches, the most of any Indian state (CPCB, October 2025), and treats about 53% of its sewage (MPCB, August 2025). Enforcement depends on sparse manual sampling.
Sensor boxes in rivers, creeks and the sea are designed to read 12 parameters every 30 seconds. Each reading is compared with the CPCB limit for that kind of location. When a limit is crossed, a case opens for an MPCB officer; AI ranks which registered industrial outfall most resembles the breach so the officer knows where to look first. The model suggests; it never establishes fault. Every case has a tamper-evident reference number and all figures are published as open JSON.
Pilot built by Aegis Knowledge Trust for evaluation by the Maharashtra Pollution Control Board. Current readings are simulated and not for enforcement. A proposal to link SaveNadi with mandatory online effluent meters (OCEMS) has been submitted to the Environment and Climate Change Department.
SaveNadi is not an official Government of Maharashtra service.
Visit SaveNadi →Students work on sensor-data pipelines, anomaly detection, source-attribution ranking and the public dashboard.
“Help a missing child return home.”
Thousands of children and women go missing in India every year. Sightings by the public rarely reach the right officer in time, and photo-matching at scale raises serious privacy and false-match risks.
Anyone who sees a child who may be missing sends a photo from their phone in under a minute: free, no login, installable as an app, and usable without a signal. The interface runs in English, Hindi, Marathi and Gujarati. Face matching proposes candidates, but a police officer checks every report before any family is contacted. The site deliberately offers no public search of people.
Live at khozo.org. Matching runs on Aarakshak with human review of every match.
Khozo helps reunite missing children, women and other missing persons with their families in India.
Visit Khozo →Students contribute to the offline-first PWA, multilingual UX, matching evaluation and officer workflow tooling.
“Sovereign verified presence for India's institutions.”
High-stakes identity decisions, exam integrity, attendance, KYC, access, need more than a match score. Institutions must show who was present, what the model decided, which evidence was reviewed and why an action was allowed.
An evidence workbench for building, running, evaluating and governing identity decisions: 1:1 face comparison, 1:N gallery search, liveness and screen-risk checks, calibrated thresholds with quality gates, a human-reviewer queue for borderline cases, and a signed evidence packet per decision linking media, scores, thresholds, timestamps, tenant, reviewer action and model version. India-hosted or on-prem, tenant-scoped storage, exportable audit packets.
Private institutional pilot; production certifications scheduled for 2027. Powers Assessfy (assessment integrity), mUniCampus (campus identity and attendance) and Khozo (missing-child sighting matching).
Research focus: making biometric AI accountable, explainable and reviewable.
Visit Aarakshak →Students work on evaluation datasets, threshold calibration, reviewer tooling and audit-packet design.
“Students learn on the systems that teach them.”
Hours of lecture video are hard to navigate, assessments are easy to game, and campus operations generate data nobody uses. Education is where Aegis can test AI with the highest standards of fairness, because our own students are the users.
Lecture-video indexing and recommendation: machine-learning models that index faculty lectures by topic, generate word clouds and content indexes, and recommend related video segments and YouTube material per subtopic. Assessment integrity: skills-based, proctored certification on Assessfy with human review of borderline cases. Campus AI: identity, attendance and exam workflows on mUniCampus, the platform that also runs Aegis admissions.
Ongoing research projects and production systems used by Aegis cohorts; see the full research archive.
The complete list of research projects, including older work, is on the Research Projects page.
All research projects →Every cohort inherits a working system and ships an improvement: better indexing, fairer assessment, faster workflows.
Government departments, NGOs, hospitals, utilities and enterprises: if you have a public-interest problem and data, Aegis Knowledge Trust and the AI Product Factory can prototype, pilot and, with you, run a system that is accountable by design.
Head – Admissions and AI Executive Education | Learning Solutions
Aegis is the AI School Partner at Bharat AI Innovation 2026, 20–21 November, World Trade Center Mumbai. SaveNadi, Khozo and Aarakshak will be demonstrated at the Aegis stand.
Tell us about your team. A program director replies within one business day.
By submitting you agree to be contacted by Aegis about research partnerships.
Your enquiry is with the Aegis corporate programs desk. We will reply within one business day. Need it faster? Call +91 7777 066 810.