Inspiration

We noticed that pharmaceutical decisions are usually driven by sales numbers and prescription data, which show what is happening in the market but not why it is happening. At the same time, patients and doctors are constantly discussing medicines online through forums, review platforms, and professional communities. These discussions reveal concerns about side effects, access issues, and whether treatments actually work in practice.

Our inspiration comes from the gap between structured business data and unstructured human experience. We want to bridge this gap by turning public discussion into a measurable signal. The goal is not to determine medical truth, but to understand how medicines are talked about and experienced across regions, and what drives those differences in perception.

What it does

Project PharmaLens will be a global web based dashboard that shows how medicines or medical services are perceived across different countries using publicly available data from patients, doctors, and institutions.

Users will select a medicine from a curated list, and the system will display a world heatmap showing how perception varies by region. By default, it will show a combined perception index, but users will be able to switch between individual views such as effectiveness perception, side effect mention frequency, access friction, trust in evidence, and competitive pressure.

The system will surface dominant themes and representative excerpts for each region so that users can understand why perception looks the way it does. The purpose is not to claim clinical accuracy, but to provide insight into how people and professionals talk about and experience medicines in real contexts.

How we built it

For this project, we will design a modular natural language processing pipeline focused on geographic and time based analysis of drug perception. Data will be collected from social media platforms, patient forums, drug review websites, doctor discussion platforms, and public clinical or institutional sources. APIs will be used where available, and policy compliant web scraping will be used for public content. Raw data will be stored in structured format so that it can be audited and reprocessed later.

The processing pipeline will clean and de duplicate text, resolve drug names to standardised identifiers, and infer country using multiple weak signals such as language, platform metadata, and location mentions. It will generate embeddings to capture topic and sentiment patterns, cluster similar narratives, and extract structured perception signals such as effectiveness, side effects, access, trust, and competition.

Aggregated metrics will be served through a backend API and visualized on an interactive world map with selectable layers and regional insights.

Challenges we ran into

One major challenge will be the noise in online data. Discussions often include misinformation, sarcasm, and irrelevant content, which makes it difficult to trust individual posts. The system will need to rely on aggregation and clustering rather than single examples.

Another challenge will be geographic inference. Most online content is not explicitly geotagged, so location will have to be inferred using indirect signals. This introduces uncertainty that must be handled carefully and communicated clearly in the visualization.

There will also be a challenge in balancing technical ambition with feasibility. While real time analysis would be ideal, batch based updates are more realistic for an early version of the system.

Accomplishments that we're proud of

We are proud of defining a clear problem that connects public perception with pharmaceutical decision making. Instead of focusing only on sentiment, we designed a framework that breaks perception into interpretable dimensions such as effectiveness, side effects, access, trust, and competition.

We are also proud of designing a system that treats uncertainty honestly, especially in geographic attribution, and makes it visible rather than hiding it. The conceptual architecture brings together data collection, language analysis, and visualization into a coherent pipeline.

What we learned

We learned that perception is not simply positive or negative. A drug can be viewed as effective but inaccessible, or trusted but associated with frequent side effects. Separating perception into multiple dimensions makes these patterns clearer and more useful.

We also learned that unstructured text can reveal meaningful market signals when processed at scale. Clustering and embeddings allow dominant narratives to emerge without relying entirely on manual labeling.

From a design perspective, we learned the importance of making complex analytics interpretable so that non technical users can still understand what is happening and why.

What's next for Project PharmaLens

Next, we will implement the full data pipeline and begin testing it on a small set of high visibility medicines. We will refine the perception dimensions based on real outputs and improve geographic inference using additional signals.

We also plan to improve the dashboard with trend analysis so that users can see how perception changes over time. In the long term, we aim to expand coverage to more medicines and regions and explore partnerships with healthcare and pharmaceutical stakeholders who could use perception data alongside traditional metrics for better decision making.