Sai Varshini Papineni

• 94% — Precision in Fraud Detection
• $2 Million — Annual Savings in Chargebacks
• 18% — Reduction in Fraud Rate

Paytm processed 500,000+ transactions daily, with rising fraud attempts. Their existing fraud detection lacked a real-time risk scoring capability and an ML-based system to accurately flag suspicious transactions at scale. High chargeback losses of $2+ million annually impacted business revenue.

SOLUTION
• Defined PRD for real-time risk-scoring engine handling 500,000+ daily transactions.
• Built fraud detection using XGBoost and TensorFlow on AWS SageMaker.
• Integrated real-time risk scoring into the merchant payments pipeline.
• Led cross-functional teams across data science, engineering, and compliance.

We achieved 94% precision in fraud detection at scale, reduced the fraud rate by 18% across merchant transactions, and saved $2+ million annually in chargeback losses using a scalable ML system that processes 500,000+ transactions per day.

• 88% — Intent Accuracy
• 30% — Ticket Deflection Rate
• 4.3 — CSAT Score

500,000+ merchants relied heavily on human support for query resolution, and high support ticket volumes led to slow resolution times. A multilingual self-serve solution to handle a diverse merchant base was lacking, and a low CSAT score of 3.6 reflected poor support experience.

SOLUTION
• Designed multilingual NLP bot using Rasa, BERT, and Dialogflow on GCP.
• Built intent classification and dialog management for accurate responses.
• Launched a bot to 500,000+ merchants across multiple languages.
• Led Agile sprints aligning NLP researchers, engineers, and support teams.

We achieved 88% intent accuracy across multilingual queries, reduced the number of support tickets deflected to human agents by 30%, and improved CSAT from 3.6 to 4.3 post-launch. We scaled to 500,000+ merchants for faster, self-serve support.

• 87% — Answer Accuracy
• 35% — Reduction in Support Dependency
• 2,000+ — Active Users in 3 Months

Students couldn’t self-serve answers from 500,000+ documents, leading to delays and overreliance on human support staff for query resolution. An intelligent system to bridge vast institutional knowledge and student needs was lacking, as was a scalable solution to handle growing query volumes efficiently.

SOLUTION
• Defined product vision and wrote technical PRDs aligning AI/ML researchers.
• Built RAG pipeline using GPT-4, LangChain, and Pinecone that indexed 500,000+ documents.
• Developed the FastAPI back end for real-time, context-aware responses
• Led Agile sprints, driving prototype to production deployment.

We achieved 87% answer accuracy across 500,000+ documents, reduced support dependency by 35%, and reached 2,000+ active users in under 3 months. We also established a repeatable model for future AI initiatives at FIU.

• 40% — Increase in Feature Engagement
• 15% — Lift in Upsell Conversion
• 10+ Million — Merchants Reached

• Merchants weren’t discovering higher-value dashboard features.
• Low feature discovery capped overall engagement.
• Cross-sell and upsell potential went untapped.
• The dashboard was functional but not driving growth.
• Any solution had to scale across 10+ million merchants.

• Identified low feature discovery as the growth bottleneck.
• Specified an AI-driven next-best-action layer.
• Surfaced tailored insights inside the dashboard.
• Ran A/B tests to tune relevance and placement.
• Scaled personalization across the full merchant base.

• Boosted feature engagement by 40%.
• Lifted upsell conversion by 15%.
• Reached 10+ million merchants with tailored recommendations.
• Turned the dashboard into a personalized growth surface.
• Guided merchants toward higher-value actions.

• 20% — YoY Growth in Transaction Volume
• 10+ Million — Year-over-year Increase in Transaction Volume

• Grew transaction volume across a 10+ million merchant base.
• Kept the core payment flow fast and reliable.
• Reduced friction from a six-step journey.
• Balanced growth with reliability at scale.
• Coordinated multiple teams on a steady cadence.

• Owned the payments dashboard roadmap end-to-end.
• Prioritized reliability and transaction-growth features.
• Redesigned journey maps from six steps to four.
• Coordinated cross-functional Agile delivery.
• Validated changes against conversion metrics.

• Drove 20% YoY transaction-volume growth.
• Reached 10+ million merchants with the redesigned flow.
• Cut the payment flow from six steps to four.
• Delivered a simpler, higher-converting experience.
• Kept the core flow fast and reliable at scale.

• 35% — Reduction in Funnel Drop-off
• 22% — Activation

• Checkout and onboarding flows were leaking users at key steps.
• Drop-off held back activation and first-transaction conversion.
• Testing had to target the highest-impact funnel points.
• Gains needed to hold up at Paytm’s transaction scale.
• Results had to be tracked on a consistent cadence.

• Built an experimentation backlog ranked by drop-off.
• Designed 20+ A/B tests across checkout and onboarding.
• Prioritized tests where funnel analytics showed biggest leaks.
• Analyzed results and shipped winning variants.
• Institutionalized a weekly metrics cadence.

• Cut funnel drop-off by 35%.
• Lifted activation rate by 22%.
• Ran a structured 20+ test experimentation program.
• Compounded gains at Paytm’s transaction scale.
• Established experimentation as a repeatable practice.

• 30% — Increase in Student Engagement
• $150,000 — Funding

• The institution lacked a reliable student-engagement data framework.
• Leaders couldn’t clearly measure how students used programs and tools.
• No shared KPIs existed to guide decisions.
• Cross-functional delivery needed structure and cadence.
• A compelling vision was needed to unlock innovation funding.

• Built a SQL and Tableau analytics framework with clear KPIs.
• Led a 12+ person cross-functional team on Agile.
• Pitched an AI product vision to leadership.
• Delivered 5+ major releases on a steady cadence.
• Kept backlog carry-over under 10% each cycle.

• Improved student engagement by 30%.
• Secured $150,000 in innovation funding.
• Shipped 5+ major releases.
• Held backlog carry-over under 10%.
• Established a reusable analytics framework with clear KPIs.

• 5+ — AI Career Tools Piloted

• Career services needed to scale personalized student guidance and matching.
• Employers needed to screen candidates faster.
• Solutions had to serve both students and employers.
• New AI tools had to be proven before scaling.
• User feedback had to drive iteration.

• Piloted AI career-matching and resume-screening tools.
• Ran workshops on AI tools like ChatGPT prep and LinkedIn AI.
• Built voice-of-customer loops with students and employers.
• Evaluated AI CRM and case-management tools for efficiency.
• Measured each tool against satisfaction and placement metrics.

• Piloted 5+ AI career tools across a two-sided base.
• Delivered faster, data-informed matching and screening.
• Strengthened student and employer engagement via VoC loops.
• Built a repeatable, feedback-led evaluation approach.

• 100% — Capability Transfer to Internal Platform Team
• 3 — Regional Sites Migrated with Zero Rollback
• 0 — Critical Incidents During Two-week Hypercare

• Acme Industrial Co. needed to migrate three regional websites from a legacy Sitecore CMS to a headless architecture (Contentful + Next.js on Vercel).
• The internal team had no prior experience operating a composable/headless platform.
• No single source of truth for identity, integrations, or governance decisions made during the migration.

• Documented the full target platform architecture across content, experience, hosting, and deployment layers.
• Produced identity and access documentation covering SSO, MFA, and a role-based access control (RBAC) matrix.
• Defined the content migration and redirect strategy.
• Compiled a go-live runbook with a phased regional cutover plan (NA → EMEA → APAC), hypercare escalation paths, and rollback procedures.
• Translated engineering and architecture decisions into documentation that non-specialists could operate from independently.

• Decision log and runbook became the operational reference for governance and incident response
• Phased go-live executed across all three regions with zero rollback triggered

• **22% ** — Prioritized Backlog to Close Difficulty Mismatch for Adult Learners
• 15% — Owned Backlog and Acceptance Criteria for The Adaptive Algorithm

Limited to only 2-3 hours of real-time overlap across all three time zones.
Content team needed pedagogical review time, delaying engineering progress.
Adaptive algorithm required stable specifications before the sprint.
Compliance risk (FERPA) needed resolution before development.
Pilot feedback loop needed to operate without delaying subsequent sprints.

Established a core standup/decision window from 11 am to 1 pm CST for cross-team overlap.
Created asynchronous handoff templates, allowing India to begin work without live approval.
Implemented a locked review cutoff for the content team before each sprint began.
Preemptively added FERPA/compliance reviews into a RAID log before development.
Ran pilot feedback in parallel batches to avoid blocking ongoing sprint work.

Reduced blocker resolution time from 2 days to under 1 day.
Achieved 90% on-time sprint delivery without full-team overlap.
Increased quiz completion rate by 22% over the baseline.
Zero compliance escalations during the pilot phase.