I build
products
that ship

Emversity was scaling from 500 to 3,000 admissions. Every additional admission required another IST. But 60% monthly attrition meant the team never actually grew — it just churned.

~40%Leads never contacted

Paid leads were going cold before any human even called. 3.5–6.7hr average TTFC. 40% of leads never meaningfully contacted. The ceiling wasn't budget — it was human throughput.

~2,307Cost/PQ (human)

3.5–6.7hTTFC

60%Monthly IST attrition

Would cost ₹2.5 Cr/month and collapse under the same attrition problem within 90 days.

Hire 250 more ISTs

Outsource to a call centre

Quality inconsistency, no CRM integration. Solved volume but not accuracy.

Designed ANIKA as a dedicated pre-qualification layer — never a replacement for counsellors, only a fast first filter. Accuracy over capability: deterministic flows, hard confidence thresholds, no hallucination risk on fee or program details.

Trade-off accepted: Gave up conversational flexibility for reliability. ANIKA can't handle edge cases — it escalates them. Intentional.

20×

Cheaper per PQ

₹391 vs ₹2,307

99.8%

Lead persistency

vs ~40% human

₹8L

20-day pilot cost

vs ₹62.5L humans

Measured over 20-day pilot, same lead source and spend.

My scope: Sole PM. Defined the architecture, wrote the PFR, owned GTM phasing, and set the confidence threshold guardrails. Engineering was 3 people.

AI · LLM · Sales Intelligence

The hardest call wasn't the AI design — it was convincing the sales head that 80% accuracy at 20× lower cost was better than 95% accuracy that would never scale.

Why this transfers: Any team hitting a headcount ceiling on outbound work faces this structural problem. AI as a capacity layer — not a replacement — applies universally.

Project 02 of 08

02 / 08AI System

AURA Intelligence Suite

AI / LLMRAGSales

Performance marketing generating leads, but only 3–5% of calls audited. Knowledge walked out with every IST who quit — and quit they did, constantly.

Three separate failures compounding: FIFO calling buried high-intent leads. QA caught nothing at scale. Top performers' best objection responses were locked in their heads, unshared.

3–5%Calls audited

FIFOLead prioritisation

₹3.75 CrRevenue at risk

Would require 8+ QA FTEs and still couldn't process 100% of calls.

Manual QA at scale

Sales script PDFs

Static knowledge ignores objection variants. Tried before — adoption near zero.

3-layer flywheel: AI Call Audit covers 100% of calls post-call. Intent Prediction reroutes high-intent leads pre-call. PitchMate RAG serves objection scripts in real-time inside the CRM iframe.

Trade-off accepted: Audit layer required ISTs to accept 100% call monitoring. Productivity dipped 2 weeks before recovering.

−28%

CAC on BVoc

vs paid baseline

−70%

Mis-selling rate

13% → 4%

+48%

IST productivity

48 → 71 calls/day

Measured over one full admission cycle (3 months) post-launch.

My scope: Sole PM. Defined all 3 modules, sequenced them (Audit first), managed IST change-management programme.

GenAI · Personalisation · Growth

Audit data made everything downstream defensible. Launching Intent Prediction without Audit first would have given better routing with zero understanding of why.

Why this transfers: Any sales team with high attrition is losing knowledge capital every quarter. The RAG chatbot pattern — capturing top performer logic, serving it in real-time — transfers to any high-variance sales motion.

Project 03 of 08

03 / 08GenAI Growth

ECHO Rendering Engine

GenAIGrowthLanding Pages

Paid acquisition was 45% of all leads. CPL had risen 25% over two cycles. Adding spend wasn't the answer — the funnel itself was leaking.

75%Post-first-scroll bounce

Every visitor saw the same generic page regardless of search intent. A student searching "OT technician Bengaluru" got the same headline as "nursing alternatives." 75% bounced after first scroll.

2.1%Visitor→lead rate

15sAvg time on page

₹946Cost per lead

More leads, worse leads. Optimising the wrong metric.

Remove form fields

Higher scholarship offers

Compresses net revenue, weakens brand trust.

Webinar-led funnels

Defers the relevance problem instead of solving it at first impression.

3-layer intent preservation: GenAI rewrites hero on utm_term in <300ms (Redis cached). Intent-mapped WhatsApp post-capture. CRM call summary so first sales call starts informed.

Trade-off accepted: GenAI in a revenue-critical path required hard guardrails — semantic deviation ≥0.82, fallback to default Atlas JSON, feature-flagged rollout.

+160%

Visitor → lead

2.1% → 5.5%

−52%

Cost per lead

₹946 → ₹445

+300%

Lead → sale

0.56% → 2.23%

Measured 6 weeks post-launch, controlled for same paid campaign parameters.

My scope: Sole PM. Framed the problem, designed the 3-layer architecture, wrote semantic guardrail spec, ran 22 bounced-user discovery interviews.

Education Operating System · 0→1

The reframe — from "low conversion" to "intent mismatch" — unlocked the entire solution space. Without it, we would have optimised buttons and form fields indefinitely.

Why this transfers: Most conversion problems in performance marketing are relevance problems in disguise. The intent-preservation architecture transfers to any funnel where user context is lost between click and contact.

Project 04 of 08

04 / 08Platform · 0→1

Hercules ERP

Platform0→1 Build13 Modules

25 campuses. 6+ disconnected tools. Finance in Tally, admissions in Excel, support on WhatsApp. Data existed but had no common backbone.

₹1.1 Cr annual revenue leaking through undisclosed discounts nobody could see. Every downstream product — AURA, emConnect, ECHO — reading dirty data. Decision-making on a foundation that didn't exist.

₹1.1 CrAnnual leakage

6+Disconnected tools

25Campuses uncoordinated

~0Data integrity

No existing ERP handled university affiliation + multi-campus + B.Voc. Customisation cost > build cost.

Buy an off-the-shelf ERP

Integrate via Zapier

Still no single source of truth. Every integration a new failure point.

Microservices-first. RBAC and Finance in v1 before any operational features — without access control and payment integrity, nothing downstream could be trusted.

Trade-off accepted: 4 months before campus managers saw any user-facing benefit. Required VP-level commitment to hold the line on timeline pressure.

₹1.1 Cr

Leakage stopped

MSP enforcement

93%

Inventory loss cut

₹7L → ₹50K/yr

+24pts

NPS lift

+18 → +42

Measured across 3 full operational quarters post-rollout.

My scope: Sole PM across all 13 modules. 40+ campus manager discovery interviews. Sequenced the build and owned rollout across 25 campuses.

Learning Management System · 0→1

Sequencing is the hardest PM decision in a 0→1 platform. Get the foundation right before adding user-visible features — even if it means 3 months of "nothing to show."

Why this transfers: Any multi-location company without a shared data layer is on borrowed time. RBAC first, single truth, microservices — same principles apply to any multi-entity platform.

Project 05 of 08

05 / 08EdTech · 0→1

Emverse LMS

LMS3-App Ecosystem0→1 Build

25+ campuses. 5,000 students. No structured learning system — just content dumps on WhatsApp and informal class delivery that varied wildly by educator.

~0Post-class retention loop

Passive content delivery was producing students who couldn't perform in clinical internships. Class time wasted re-covering basics. No visibility into whether students had prepared. Post-class retention near zero.

0%Structured pre-class

Ad-hocClassroom delivery

WhatsAppPrimary content channel

Buy an off-the-shelf LMS (Moodle, Canvas)

None supported the pre/in/post class loop, OSCE practicals, or the TV App classroom surface. Customisation cost exceeded build cost.

Video-only content platform

Solves content delivery, not learning. Without EmBot, class overview, and reflection — it's passive consumption with a better UI.

Designed EWL (Emversity Way of Learning) framework first — pre/in/post class pedagogy — then built three apps to serve it: Student App for self-study and reflection, Tutor App for structured classroom facilitation, TV App for shared classroom display. EmBot as a subject-scoped AI doubt solver that feeds tutor insights before class.

Trade-off accepted: Rejected generic AI chatbot in favour of subject-scoped EmBot. Lost flexibility, gained trust. Students ignored generic AI — they used the Biochemistry doubt solver.

4.8

Student NPS

highest at Emversity

5,000

Daily active users

100% of enrolled students

100%

Internship placement

VR + OSCE prep cited

3× engagement improvement vs. prior delivery model. Measured over one full academic cycle.

My scope: Sole PM across all three apps. Defined EWL framework, wrote PRDs for all 12 features, owned tutor change management across 25 campuses.

Affiliate Acquisition · PLG

The hardest problem wasn't the Student App — it was getting 100+ educators to change how they run a classroom. Change management cost more time than engineering. Every campus had a different informal workflow I had to understand before I could replace it.

Why this transfers: Any education product without a pedagogical framework is just a content player. The EWL pre/in/post loop — preparation, facilitation, reflection — is a transferable pattern for any domain where passive consumption doesn't produce skill.

Project 06 of 08

06 / 08PLG · Mobile

emConnect

PLGMobile 0→1Founding PM

Paid CAC was ₹70–90K per admission. 40+ teacher interviews revealed hundreds informally referring students for free — no visibility, no payout, no infrastructure.

A distribution channel already existed in the wild — decentralised, informal, unmonetised. Teachers referring 4–5 students/year with zero recognition. The channel had latent demand. It had no product.

₹70–90KBlended paid CAC

0Structured infra

40+Teacher interviews

100%Referrals untracked

Manual incentive programme

No KYC, no attribution, rampant fraud. Payout disputes killed trust within weeks.

Add referral tab to LMS

Wrong surface. Affiliates are external consultants, not students. Needed standalone identity.

Complete affiliate lifecycle platform. Core principle: trust is the product. KYC verification, real-time lead status, sub-24hr payouts. In-product referral loop: refer an affiliate who gets 3 admissions → earn ₹10K bonus.

Trade-off accepted: Deferred gamification and multi-level network to v2 to protect v1 stability. Would have added 6 weeks and fraud surface area.

10K+

Affiliates onboarded

in 4 months

44%

Via PLG loop

zero team intervention

₹17.9 Cr

Gross revenue

₹17.11 Cr net

CAC via emConnect: ₹20–25K vs ₹70–90K paid. Became 2nd largest channel after Google Ads.

My scope: Sole PM. 40+ field interviews, full platform architecture, PLG mechanic design, cross-functional execution with sales, finance, engineering.

Visibility drove 3× more affiliate participation than payout amounts alone. The moment affiliates could see their live pipeline — referral behaviour changed completely.

Why this transfers: The best distribution channels are hiding in plain sight. The PM's job is to find them, formalise them, and remove the friction keeping them small.

Project 07 of 08

07 / 08PLG · Growth

Referral System

Zero-CAC Growth Engine

PLGGrowthZero-CAC

High student satisfaction with no structured mechanism to turn it into referrals. Every referral happening organically — untracked and unrewarded.

NPS +18Before system launch

Student referrals visible in NPS but invisible in data. No attribution, no reward loop, no product infrastructure. The behavioural signal was strong. The product was zero.

0Tracked referrals

HighOrganic word-of-mouth

₹0Referral revenue

Early tests: cash alone produced one-time referrals. No compound behaviour.

Cash reward only

WhatsApp broadcasts

Spam-adjacent, low trust, no attribution. Damages brand among advocates.

Dual-engine referral: CRM-driven (triggered at journey milestones) + product-driven (gamified tiers in LMS/ERP). Key insight: making progress visible drove 3× more referrals than cash alone.

Trade-off accepted: Gamification added product complexity and abuse-prevention logic. 3 extra weeks. Worth it — tier visibility was the core mechanism.

₹23 Cr

Referral revenue

zero additional CAC

446

Admissions driven

via referral channel

−35%

Blended CAC

across all channels

Measured over full programme lifetime. Became 2nd largest acquisition channel.

My scope: Sole PM. Designed dual-engine architecture, built gamification tier logic, ran discovery to identify peak referral-intent moments.

B2C Psychometric → B2B2C SaaS Pivot

Visibility compounds participation. A progress bar or tier badge changes referral from transaction to relationship.

Why this transfers: Any product with genuinely satisfied users is sitting on an untapped acquisition channel. The question is whether you've built infrastructure to make that satisfaction visible.

Project 08 of 08

08 / 08Pivot · 0→1

Emerge → 42.ai

PivotB2B2CAI

₹4.25 Cr in B2C psychometric revenue. High satisfaction scores. But unit economics were broken — high CAC, low LTV, no retention loop.

5-failure RCA was definitive. All five failures shared the same root: unit economics of B2C student psychometrics in India don't work.

5 simultaneous structural failures: no retention loop, high CAC vs low willingness-to-pay, purchase authority with parents not students, no distribution leverage, no data moat. Not one problem — five, compounding.

Add retention features

Retention wasn't the core problem — distribution and WTP were. More features on a broken model is expensive noise.

Pivoted to B2B2C. Schools as distribution partner. Same RIASEC scoring engine, rebuilt interface: school dashboard, student report, Zuno parent view, counsellor workspace, school heatmaps.

Trade-off accepted: Sunsetting ₹4.25 Cr of existing B2C revenue. The 5-failure RCA made it defensible — data-backed, not a gut call.

₹4.25 Cr

B2C revenue

gracefully graduated

5/5

RCA failures resolved

in B2B2C model

100+

Schools targeted

B2B2C pipeline

B2B2C launch in progress. B2C wound down over 2 admission cycles.

My scope: Sole PM. Ran the 5-failure RCA, made pivot recommendation to leadership, designed B2B2C architecture, led re-engineering of school-facing interface.