A founder from Silicon Valley came to us with a product that was already working. The AI logic was solid. The backend did what it was supposed to. DataScoop could analyze metric changes, find the root causes and explain them in plain language - which, honestly, is genuinely hard to build. The problem was everything the user actually saw. The interface hadn't kept up with the product. There were Figma designs ready, a completely new visual direction, but no one had translated them into a working frontend yet. The existing code wasn't built to carry them. And the data visualization layer, the part that makes analytics actually readable, needed to be rebuilt from scratch. So the ask was clear on the surface: rebuild the frontend, implement the new design, make it fast. In practice, it was more nuanced than that. Complex data files needed to be parsed and turned into charts that told a story, not just displayed numbers. Performance had to hold up so that insights landed in seconds, not after a loading spinner that made you forget what you were looking at. And all of it had to feel like it belonged together: the AI output, the visualizations, the interface, the Slack and email delivery. We rebuilt the frontend in React.js and TypeScript, applied strict code quality standards from the start and worked through the Figma designs screen by screen, not just implementing them, but pressure-testing how they behaved with real data. The hardest part wasn't any single technical problem. It was maintaining coherence, making sure that as the pieces came together, they felt like one product, not a patchwork. The result is a platform where a data analyst can open it, see a metric move, and understand why, in plain language, in seconds, delivered wherever they actually work. That's the version DataScoop's users have now. Check it out

Great job

Project: Velocity SaaS – Revenue Turnaround Analysis Type: Independent portfolio project Dataset: Synthetic B2B SaaS, 2,000 accounts / 2.2M rows, benchmarked against ProfitWell/OpenView

What I did:

MRR waterfall analysis, churn segmentation by plan/industry/channel

Usage decay / silent churn risk modeling

Logistic regression churn predictor: 98% recall, 70% precision, 0.82 F1

Built executive Looker Studio dashboard + Streamlit churn predictor

Outcome: $411K silent MRR at risk identified, CS-ready prioritized account list Stack: SQL, DuckDB, Python, Pandas, Scikit-learn, Looker Studio, Streamlit

Live app: velocity-saas-churn-predictor.streamlit.app Repo: github.com/SURAJRAJPUT2006