Institute of Innovation · PhysicsWallah
Turning scattered, manual, "message-someone-to-find-out" problems into working software — across six PW verticals.
At a glance
We embed with senior stakeholders across the org, take an ambiguous ask, and hand back a working system — from an admissions analyst-in-a-box to a national tender-watch engine.
The footprint
Answers lived across 9+ scattered sheets. Getting one meant hand-digging or messaging someone — and there was no forecast leadership could rely on.
A login-gated platform: staff ask in plain English, an AI agent writes read-only SQL over a live copy of the data → exact answers, tables, charts. A 205-table analytical store over 9 live sources, migrated onto PW's own cloud.
Used daily by IOI admissions leadership. Same question → the same audited number, every time — so a table that took an analyst a day is now a 20-second question.
A deterministic 3-method ensemble (no black box), validated by replaying the engine against history — every number traces to its basis.
An honest end-of-cycle projection with uncertainty bands, a per-campus "funnel breaks" diagnostic that shows exactly where students are lost, and automated daily email reports — the forecasting story the team referenced icedoutai for, built and validated in-house.
Onboarding ran on a 14-tab manual sheet. Academic fees are paid to Medhavi — not PW's ERP — so a student's real payment status was split across ERP, the Medhavi sheet and loans, with no single truth.
A campus-wise portal, live from all three sources, that reconciles each student by identity-matching (ID → email → mobile), with role-based access for finance / campus-admins / counsellors and physical onboarding tracking. Migrated to the cloud and made 20–30× faster.
Live on the 2030 cohort — 268 students across 8 campuses — turning a manual, error-prone sheet into a real-time, single source of payment truth.
The govt team hand-checked GeM plus dozens of state and central portals — slow, and blind to tenders hidden behind generic titles like "Custom Bid for Services."
An engine that scrapes GeM + ~38 portals daily, AI-rates every tender for PW fit, and reads inside the bid PDFs (OCR for scanned & Hindi) to pull structured requirements — scope, eligibility, EMD, dates. Semantic search surfaces high-fit tenders the title can't reveal. Scored against PW's real bid history.
A live watch over ~190,000 tenders (incl. ~68,000 GeM bids) across ~38 portals a day. The vertical's manager greenlit extending it toward AI-assisted bid responses.
The Aarambh event registration page (Bengaluru / Noida / Deoria) and AI Startup Challenge deadline management — merged into PW's real frontend codebase through the same review pipeline as full-time engineers.
A unified registration + Razorpay payment flow to replace a leaky 3-step funnel (LinkedIn form → Google Form → manual call → payment link) with one embedded form — built and passing, live once payment credentials land.
Closes a lead-capture gap that was silently losing prospects — and every unpaid registration now automatically becomes the sales follow-up list that never existed before.
A 10-stage pre-placement process across 5 teams ran on WhatsApp, calls and sheets — status invisible, handoffs lost. Designed a read / visualize / nudge layer over the existing sheets: an automatic email reminder engine + a 4-screen dashboard tracking every batch across all 10 stages — so no team changes how they work. Build underway.
Scoped an AI system to monitor the top 200 target companies — and flagged the DPDP / data-privacy line before any build.
Captured requirements for holistic student profiling beyond marks, using only existing data — no new burden on students.
The impact, consolidated
Used daily by IOI admissions & finance leadership; reviewed and extended by vertical managers in Government and Startups. Every number reflects systems and data on record.
The through-line
Divyam Jindal · Arnav Sharma — IOI AI Innovation, PhysicsWallah