In-App Analytics

The product showed analytics to teachers, schools, and districts—but the architecture was inefficient (one massive query per page returning JSON for every chart) and the metrics weren't always clear.

My background as a data visualization analyst meant I could serve as a bridge between product, engineering, and the data. I advised on what charts to use, how to compose metrics, how to differentiate them from each other—bringing both technical skills and a deep interest in effective visualization. Helped migrate from slow MongoDB queries to efficient Snowflake queries by building proper tables, eventually served via the D.R.E.A.M. API (Data Rules Everything Around Me 👐🏻) or direct stored procedures.

Broke up monolithic queries into per-chart requests that could actually scale. The performance improved. The architecture made sense. Engineers could extend it without understanding the entire system.

Custom Partner Reporting

Our largest enterprise partner (200+ schools) had specific needs—regional leaders who needed to see their cluster's data without accessing others'.

I collaborated with the Partner Success team to build a custom Tableau dashboard. Designed it so they could manage and extend it themselves—no technical background required, no waiting on a data person for every request. The same approach worked for other partners with similar needs.

This pattern—build once, hand off, enable the team—defined how I approached custom work. The goal isn't to be indispensable. The goal is to build something that works without you.

Internal Analytics

I ran workshops helping non-technical stakeholders answer their own questions. The goal was always empowerment—giving people the tools and context to explore data without waiting on the data team.

Empathy Through Data

I built dashboards that showed Partner Success exactly what partners see in the app—same metrics, same views. The goal was empathy: helping them put themselves in their partners' shoes. Alongside that, I added enriched context—engagement trends, comparative analysis, operational metrics—so they had the fuller picture partners don't see.

The Self-Service Journey

Over four years, this evolved. Started with internal dashboards in Snowflake using their built-in dashboards feature. Then Tableau. Then Streamlit apps for things that needed custom logic. Then we got Sigma, which let people build their own dashboards. Now we have both Streamlit and Sigma—different tools for different needs. Created org-wide metrics dashboards for leadership along the way.

AI-Powered Analytics Foundation

At the end of my time there, I built an internal MCP server so product, partner success, and engineering could query Snowflake safely through Claude—letting them audit metrics, explore data, and ask questions without waiting on me.

The elegance is in reusing what we already had. The MCP server queries Snowflake directly, enabling real action. Tools like describe table and search catalog are fed by artifacts that dbt automatically creates—technical documentation that lives as JSON in the repo, now exposed in a user-friendly way. Claude can explain things in natural language, with the user's context. The server also includes prompts for frequently asked data questions. No new infrastructure—just existing assets made accessible in new ways.

Beyond the MCP server, I ran workshops for product and engineering on how to leverage AI tools for data needs—custom GPTs, ChatGPT, Claude Code, GitHub Copilot in VS Code. Teaching people to access and work with data using these tools, so they didn't have to wait on the data team.

This was the first step toward something bigger: analytics where the interface is conversation, not dashboards. The foundation for what comes next.

Impact

• Partner Success empowered—dashboards they could manage and extend themselves, designed for handoff
• Self-service evolution—from Snowflake dashboards to Tableau to Streamlit and Sigma, meeting people where they are
• In-app analytics scaled—migrated from monolithic queries to per-chart architecture via the D.R.E.A.M. API
• MCP server—AI-assisted data exploration using existing dbt artifacts
• AI workshops—teaching product and engineering to leverage Claude, GPTs, and Copilot for data needs

Technologies