BloomSync AI

To be completed## About the Project: BloomSync AI

Inspiration 🌱

Apple trees typically bear fruit only on old wood, making mechanization and uniform fruiting difficult. We were inspired by the vision of the Farm of the Future, where genetics, AI, and automation work together to redesign crops for robotic harvesting. Our goal: to use bioinformatics + reinforcement learning to reprogram flowering genes—so apples can bloom and fruit on new wood.

What We Built ⚙️

We created BloomSync AI, a pipeline that:

Aggregates genomic data from NCBI, UniProt, and GDR for Malus domestica (apple).
Identifies and aligns key flowering genes (e.g., ( FT ), ( TFL1 ), ( AP1 )) across species.
Uses a reinforcement learning agent to explore and propose edits that increase FT-like activity.
Incorporates a deep neural reward model that evaluates each edit based on:
$$ R = w_1 S_{FT} - w_2 S_{TFL1} - w_3 N_{edits} + w_4 V_{motifs} $$ where $S_{FT}$ and $S_{TFL1}$ are similarity scores, $N_{edits}$ is the number of mutations, and $V_{motifs}$ ensures biological feasibility.

What We Learned 📚

How to navigate and mine biological databases (NCBI, UniProt, GDR) for plant genomics.
How reinforcement learning can be adapted to sequence optimization problems.
The complexity of translating bioinformatics insight into AI-driven decision policies.
That biology and computer science can be tightly integrated through representation learning.

Challenges We Faced 🧩

The massive search space of possible edits (~$4^n$ DNA combinations).
Designing a reward function that balances accuracy, biological plausibility, and minimal edits.
Interpreting real genomic annotations with limited wet-lab validation data.
Achieving a balance between scientific realism and hackathon feasibility.

Takeaway 🌍

BloomSync AI demonstrates how AI-guided genome editing could accelerate breeding programs. Our model offers a preview of a future where reinforcement learning agents propose genetic improvements autonomously—unlocking earlier fruiting, higher yield, and robotic-friendly orchards.