Data Science Analyst

Thermo Fisher Scientific · 05/2024 – Present

ML Forecast

Forecast dashboard: weekly index by fiscal week with forecast, risk levels by type, and a US map of the latest week
What it is
A rolling 13-week forecast covering multiple financial and product-utilization metrics, broken down by type and refreshed every week.
How I built it
Brought environmental factors in alongside historical trends, combining machine learning with classical econometrics. The forecast holds 90%+ accuracy.
What the dashboard shows
Compares this year against past years by fiscal week, flags each type’s risk level (low → very high) for the weeks ahead, maps the most recent week’s risk across the US, and tracks YTD growth and risk by fiscal period.
Impact
Drives weekly planning and budget conversations.

Market Opportunity

Market Opportunity deliverables: an interactive ZIP-level map with a hover panel showing sales, predicted sales and sales opportunity, plus the same data in Excel
The data
Gathered sales, demographic (population, income, insured %) and environmental (pollen) data and structured it monthly at the ZIP-code level.
How I built it
Built and compared prediction models: gradient-boosted trees, an ensemble and a winsorized statistical model. Prescriber (NPI) density and population turned out to be the two strongest indicators of sales.
What it predicts
Scaled from regional to nationwide projections. It predicts sales for ZIPs with no prior sales and flags underperforming ones, so predicted minus actual becomes the sales opportunity.
What leadership gets
An interactive map in a single HTML file: hover over any ZIP to see sales, predicted sales, opportunity, NPI count, pollen, population, income and more. Plus an Excel export that can be uploaded to Unity Catalog.

Research on Top Factors Impacting Sales

Top factors impacting sales: feature importance table and a bar chart of normalized model gain importanceCross-quarter correlation heatmap of normalized features, showing no impact in Q1, moderate in Q2, low in Q3 and heavy in Q4
The scope
Looked at every US ZIP code with sales and compared it against every available environmental and demographic factor.
How I measured it
Used feature-engineered econometric and tree-based models to measure how much each factor moves sales, ranked on a 0–100 importance scale.
What it found
Cross-quarter correlation showed the effect is seasonal: little to none in Q1 and Q3, moderate in Q2 and heavy in Q4, on both volume and utilization.
Impact
Turned the strongest drivers into strengths for the go-to-market strategy.

Leading with AI

Unlocking Productivity with ChatGPT Enterprise: Gen2Gen BRG ChatGPT training series, with sessions on the transition to ChatGPT, GPT agents, emails, docs and PMD, Excel and routing
My role
As IDD’s Gen AI Ambassador, I partner with digital literacy managers to train global teams on Gen-AI tools.
The training series
Lead “Unlocking Productivity with ChatGPT Enterprise” with the Gen2Gen BRG: the transition to ChatGPT, GPT agents, emails, docs & PMD, and Excel & routing.
What I build
Internal GPT agents used by multiple teams.
Impact
The agents save 10–12 hours a week across those teams.