I'm Dona. I build analysis that survives scrutiny, not just charts that look clean at first glance. SQL, Python, and BI dashboards, applied to problems in finance, healthcare, and product, with the assumptions checked before the recommendation goes out.
My core work is SQL, Python, and BI dashboarding, Power BI and Tableau for the visual layer, PostgreSQL underneath. I've worked across financial sentiment analysis, fund overlap modeling, checkout experimentation, and clinical trial data, always with the same question: what should someone actually do with this?
I graduated with a B.E. in Electronics & Computer Science from SIES Graduate School of Technology, Mumbai University, in May 2026.
Tested a one-page checkout against the existing multi-step flow across 42,000 users over 21 days. The headline number looked clean, but it needed to survive scrutiny before it could justify a launch decision.
Ran a randomization-health check that caught a mix-shift bug: the mobile app had shipped the treatment flag early, over-representing mobile and new users who convert better regardless of arm. Segmented by device and user type to find the real effect, then ran a power check to separate "no effect" from "inconclusive."
Found a genuine +5.2pp lift on mobile and new users, significant, while desktop and returning users were underpowered, not proven zero. Also flagged a real ~10.6% page-load regression on treatment. Recommendation: ship on mobile, hold desktop for a cleaner re-test, fix the flag-rollout bug.
Financial sentiment lives in scattered news and text, disconnected from the live price action it is supposed to explain.
Combined FinBERT for sentiment scoring, Groq/Llama for summarization, and yfinance for live market data into a single Streamlit tool that reads news and price movement together instead of separately.
A working tool that pairs sentiment signal with live market data in one view, instead of requiring two separate lookups.
Investors often hold several "different" mutual funds without realizing how much their underlying holdings overlap, quietly concentrating risk instead of spreading it.
Built ISIN-based matching logic across HDFC Large Cap, HDFC Small Cap, and Mirae Asset Flexi Cap holdings, then visualized the overlap in an interactive Tableau dashboard so a user could inspect any fund pair directly.
A dashboard that makes hidden overlap between funds visible and inspectable, instead of assumed.
Dropout patterns across clinical trials sit scattered across individual ClinicalTrials.gov listings, making cross-study comparison slow and manual.
Built a pipeline pulling and processing trial records through the ClinicalTrials.gov v2 API, structured for comparison across studies rather than one at a time.
A structured dataset that turns scattered trial listings into something comparable at a glance.
Real-time fire detection on a hexacopter cannot depend on a live connection. It has to run entirely on the hardware it is carrying.
Trained and deployed MobileNetV2 for real-time fire detection, running fully on a Raspberry Pi 5 onboard the vehicle.
96.2% detection accuracy, running entirely on-device with no cloud dependency.
Looking for a Data Analyst or Business Analyst who is comfortable with SQL, dashboards, and the messy work in between? Tell me about the role.