Inspiration
We were inspired by research highlighting the prevalence of LASA medication risks in India and the challenges pharmacists face when prescriptions primarily contain brand names.
What it does
LASA Guard is an AI-powered medication-name confusion detection and risk-assessment platform focused on Indian pharmaceutical brands.
Its key capabilities include:
Medicine Name Detection: Accepts medicine names through manual input, with prescription image and voice input as planned extensions.
Multi-Dimensional Similarity Analysis: Uses spelling, phonetic, character-pattern and embedding-based similarity techniques to identify potentially confusing brand names.
Composition Comparison: Compares active ingredients, formulations and therapeutic classifications to identify important differences between similar-sounding medicines.
Explainable Risk Assessment: Combines similarity signals and medicine information to generate interpretable risk alerts.
LASA Intelligence Dashboard: Ranks potentially confusable medicine pairs and presents insights into medicine-name confusion patterns within the collected dataset.
The system acts as an early-warning and verification-support tool, not a replacement for professional medical judgment.
How we built it
We designed LASA Guard around a hybrid AI/ML pipeline.
Data Collection: Gather medicine-brand information, generic names and composition details from accessible pharmaceutical datasets and published research.
Data Preprocessing: Clean, normalize and structure medicine names while preserving their original forms.
Similarity Detection: Apply Levenshtein Distance, Jaro-Winkler Similarity, phonetic matching and character n-grams to identify potentially confusing names.
Embedding-Based Comparison: Generate vector representations and use cosine similarity to explore additional name similarities.
Composition Analysis: Retrieve and compare active ingredients and therapeutic classifications.
Risk Scoring: Combine the different similarity and composition signals into an explainable risk-ranking model.
Visualization: Present potential confusion alerts and ranked LASA pairs through an interactive dashboard.
Our proposed technology stack includes Python, FastAPI, React.js, PostgreSQL, Scikit-learn, RapidFuzz, Sentence Transformers and FAISS.
Challenges we ran into
One of our biggest challenges was pharmaceutical data availability and reliability. A comprehensive, verified dataset connecting Indian medicine brand names with their active ingredients and compositions is essential for meaningful detection.
Accomplishments that we're proud of
We are proud of developing a solution around an important but relatively underexplored healthcare safety problem.
What we learned
Through this project, we explored how traditional algorithms and modern machine-learning techniques can complement each other.
What's next for LASA Guard
Expanding the database with verified Indian pharmaceutical brand names and composition information. Integrating accessible government pharmaceutical data sources and APIs. Introducing OCR-based prescription scanning to extract medicine names automatically. Adding voice-based input to support spoken medicine-name identification.
Log in or sign up for Devpost to join the conversation.