About AskPharma.ai

About AskPharma.ai

Welcome to AskPharma.ai, your trusted companion in exploring the world of FDA-approved medicines. Created by Harsh Singhal, AskPharmaAI simplifies complex pharmaceutical information and makes it accessible to professionals, students, and general population alike.

Our Vision

Pharmaceutical research and learning are often hampered by fragmented resources, information asymmetry, and time-consuming workflows. My vision is to bridge these gaps, ensuring that essential information about medicines is readily available, transparent, and easy to understand.

Our Technology

• FDA Drug Labeling Dataset: We use publicly available data from the FDA, including comprehensive drug labeling information, adverse reaction records, and more.
• Llama 3.3 70B Large Language Model: This state-of-the-art AI model enables us to analyze, summarize, and present data in a user-friendly manner.
• Smart Search and Retrieval: Our application retrieves data from multiple sources, optimizing your learning and research workflow.

Finding Similar Drugs

One of the standout features of AskPharma.ai is its ability to find similar drugs effortlessly. We achieve this by leveraging advanced vector cosine similarity techniques on embeddings generated from the Summary field of each drug. Here's how it works:

• Embedding Generation: We use pre-trained embedding models to create dense vector representations of the Summary field for each drug.
• Similarity Measurement: The cosine similarity between vectors is computed to determine how closely related the drugs are.
• Result Optimization: This allows us to display a list of drugs that are highly similar in terms of purpose, mechanism, or usage, aiding in comparison and discovery.

This feature empowers users to explore related drugs effortlessly, making research and learning more efficient and insightful.

Summarization using LLAMA 3.3 70B Model

We leverage the powerful LLAMA 3.3 70B model to parse and summarize data from the FDA Drug Labeling dataset (open.fda.gov). Our summaries encompass:

• Clinical information
• Dosage details
• Considerations across special populations

Note: We currently process approximately 15,000 entries based on our data completeness score. While the complete FDA dataset contains over 200,000 entries, we've optimized for performance and reliability in our cloud deployment.

Search Implementation

We utilize Whoosh as an embedded search index to power our search functionality. The index covers multiple fields including:

• Drug names
• Generic names
• Overall summary details