Piyush Agrawal

• 35% — Lower Error Rates
• 3X — Faster Onboarding for New Integrations
• 60% Reduction — In time-to-file

• Manual compliance reviews, inconsistent ERP data mapping, and frequent edge-case errors slowed scalability.
• Needed intelligent automation that could adapt to dynamic tax rule changes and multi-system dependencies.

• Mapped end-to-end workflows to identify automation opportunities.
• Designed modular microservices and event-driven pipelines for scale.
• Integrated LLMs + RAG models for anomaly detection and support copilots.
• Created explainable AI layers and audit logs to ensure trust and compliance.

• Achieved enterprise-grade automation and accuracy benchmarks.
• Improved customer satisfaction (NPS +40%) and retention.
• Established Galvix as a credible, AI-first alternative to legacy compliance systems.

• 100X — Increase in Organic Traffic
• 10,000+ — Pages Generated Programmatically Using Structured Data
• 90+ — Lighthouse SEO Score

• Galvix had limited inbound visibility and needed scale without manual content ops.
• Tax data complexity made consistent metadata and localization difficult.
• Tax data kept changing with new rules, content had to be continuously updated to keep it fresh and accurate.

• Designed a data-to-content pipeline combining tax data, NLP models, and metadata generators.
• Implemented LLM-based summarization, FAQ synthesis, and schema markup automation.
• Built validation scripts, content scoring, and sitemap auto-generation workflows.

• Became a top-ranking site in hundreds of niche tax queries.
• Reduced CAC dramatically through organic inbound growth.
• Created an extensible AI-driven SEO system adaptable across verticals.

• 18% — Improvement in Weekly User Retention
• 300+ Million — MAUs Positively Impacted via Safer Search
• 40% — Reduction in Harmful Content Exposure

• Reddit’s massive user base required scalable experimentation and data-driven decisions to enhance engagement while minimizing exposure to harmful or NSFW content.
• Needed faster insights and better instrumentation to optimize user safety and search experience.

• Designed and ran A/B experiments using internal frameworks across web and mobile, defining north-star metrics for retention and engagement.
• Partnered with ML engineers to deploy NLP-based content classifiers and automated Safe Search filters.
• Created end-to-end event pipelines for funnel visibility, integrating BigQuery, Amplitude, and internal data APIs.
• Built governance around experiment analysis, ensuring statistically sound decisions.

• Improved platform-wide safety and trust while maintaining engagement growth.
• Proved ability to merge AI, behavioral analytics, and product design to drive measurable impact at scale.

• 30% — Reduction in Shipment Processing Times
• 25% — Increased Logistics Throughput and Partner Satisfaction

• Janio operated in a fragmented logistics ecosystem across Southeast Asia with minimal data visibility, manual routing, and disconnected carrier systems.
• Needed a scalable, unified SaaS platform to integrate partners and automate fulfillment workflows.

• Built Janio’s first AI-driven digital freight platform, using predictive modeling for routing and demand forecasting.
• Defined core product architecture, APIs, and analytics stack supporting multi-country logistics orchestration.
• Introduced product management best practices—OKRs, roadmap governance, feedback loops, and Agile delivery.
• Led partnerships with top eCommerce players (e.g., Bukalapak, Zilingo, Under Armour) and built robust integrations.

• Transformed Janio into a data-driven cross-border eCommerce logistics platform, reducing cost-to-deliver and increasing visibility.
• Established scalable architecture supporting future ML routing and real-time optimization.

I led the product design and implementation of SixthSense, an RFID-based enterprise intelligence system at Microsoft.

This system combined mobility information obtained from RFID-based sensing with information from enterprise systems such as calendar and presence to automatically draw inferences about the association and interaction amongst people, objects, and workspaces. For instance, SixthSense was able to automatically distinguish between people and objects, learn the identities of people, and infer the ownership of objects by people.

As part of SuperProfs, a key challenge was digitizing the classroom content of our partner professors. We initially asked the professors to record their lectures at our state-of-the-art recording studios, but it became clear that this wouldn’t work. It’s an asset-heavy model, and it wouldn’t scale as we grew.

We needed an extremely easy-to-use, fully automated lecture capturing solution that worked with the existing infrastructure of Indian classrooms. Indian classrooms are often plagued by poor internet connectivity, a lack of a consistent power supply, a lack of trained A/V technicians, and more. I came up with the idea of using off-the-shelf Android phones with high-resolution cameras for lecture recording.

With a team of four engineers and a designer, we built an Android app and a cloud-based admin system that could automatically start/stop the lecture recording on remote Android devices. We also implemented the capability to track the professor in the classroom as they moved around using computer vision and face recognition techniques.

This model enabled us to scale to hundreds of classrooms in an extremely cost-effective, yet robust manner—becoming a key selling point of SuperProfs platform for professors in India.

Tasks Accomplished
• Designed and implemented a video streaming protocol for wireless networks which reduced packet loss by 45% and improved received video quality by 2-10 decibels in PSNR.
• Presented this work at the ACM Multimedia conference in Germany and won the Best Student Paper Award.