Genomic Counseling AI

Genomic Counseling AI empowers anyone to unlock their genetic data: instantly compute disease risks, understand them with generative AI, and receive actionable health insights in minutes.

Comment

Inspiration

Genomic testing has massive potential, but for most people, DNA data remains a confusing wall of code. I was inspired by the gap between raw personal genomics and real, actionable health advice.

Traditional genetic counseling is expensive, slow, and often inaccessible. With Genomic Counseling AI, I set out to build an intelligent, automated solution that feels like a virtual counselor and brings clarity to anyone, anywhere.

What it does

Genomic Counseling AI turns raw DNA into clear, personalized insights.

Users upload a TXT genotype file. The system analyzes genetic markers, detects disease-linked variants like APOE ε4 for Alzheimer’s, and calculates polygenic risk scores across thousands of traits using real GWAS data.

Using Amazon Bedrock, it then generates a structured, human-readable report that explains risks and gives evidence-based health recommendations. The experience mirrors a real genetic counseling session - automated, accurate, and empowering.

How we built it

The frontend was built in Streamlit, offering a clean, single-page app for uploading genotype files.

Under the hood, I developed a Python pipeline that detects disease-associated SNPs using a curated adult-onset trait catalog and computes polygenic risk scores with simulation-based percentiles for each trait.

The generative layer uses Amazon Bedrock to power an LLM that converts raw statistical output into a readable, user-friendly interpretation. Prompts were designed carefully to ensure medical responsibility, tone, and clarity.

I used pandas, NumPy, and statsmodels for data processing, while AWS SDKs handled secure model inference with Bedrock’s Amazon Nova foundation model.

Challenges we ran into

Parsing diverse genotype formats was tricky. Variations in naming, encoding, and missing data demanded robust preprocessing and validation.

Cleaning the GWAS catalog data was equally tough. Normalizing effect sizes, handling missing SNPs, and ensuring chromosomal consistency took significant effort.

Simulating PRS distributions per trait required careful statistical handling and filtering for reliability. Matching the user’s score to population percentiles had to be both fast and accurate.

Integrating generative AI also posed challenges. Prompts had to be precise - scientifically grounded but readable, with zero room for medical ambiguity. It took several iterations to get the balance right.

Accomplishments that we're proud of

Before this hackathon, I had never worked with genomic data. As an Alzheimer’s researcher and MSc Electronic Engineer, this was a jump into the deep end - and I came out with a working product.

I learned to process raw genotype files, compute accurate PRS using real GWAS data, and simulate population scores to generate personalized percentile reports.

I also mastered AWS tools for the first time, integrating Amazon Bedrock to generate AI-driven reports from raw genetic input. Every output respects both privacy and medical relevance. In just a few days, I built an end-to-end genetic counseling experience - automated, secure, and deployable anywhere. It’s not just a hackathon win. It’s a blueprint for democratizing precision health.

What we learned

I learned how to handle genomic data from scratch - how to parse genotypes, match SNPs, and generate meaningful outputs using PRS and single-variant associations.

I also dove deep into AWS, learning to use Amazon Bedrock, structure LLM prompts for sensitive healthcare data, and build reliable inference pipelines with the AWS SDK.

Designing interpretable, actionable AI outputs taught me the power of prompt engineering in healthcare.

Most importantly, I learned how to connect multiple worlds: genomics, cloud AI, engineering, and UX. That blend is where real innovation happens.

What's next for Genomic Counseling AI

Next, I’ll refine the UI/UX for even better accessibility. I’ll expand the catalog with more traits, conditions, and ancestry insights, extending beyond health.

I plan to test other LLMs, like Biomistral on AWS, to boost accuracy and domain-specific reasoning. Ultimately, I want to turn this into a robust, inclusive genomics platform - empowering everyone to understand their DNA.

Built With

Share this project:

Updates

Submission history