DSTA BrainHack TIL-AI 2025

Team: Hackoholics

👥 Team Members

Name	Role
richierx	RL, ASR
w3joe	RL
cjiayu03	OCR
electrum21	CV

We managed to qualify with 13th place in the Novice category, but was defeated in the semi-finals.

---

Contents

• Introduction
  • RL
  • ASR
  • OCR
  • CV
  • Get started
  • Understanding this repo
  • Build, test, and submit

RL

In a multi-agent environment, the Scout must navigate a dynamic map, avoid capture by the Guards, collect Recon Points, and complete challenges at fixed locations. Meanwhile, the Guards are trained to patrol and intercept the Scout efficiently.

Solution:
We implemented a hybrid approach: supervised learning and imitation learning were used to teach the Scout to move strategically, while the Guards were trained using a CNN-based DQN agent with experience generated via BFS pathfinding.

📄 Read more about RL →

ASR

The challenge is to transcribe spoken content from noisy audio recordings. The model must handle speech under varying conditions and accurately produce a written transcript despite distortions or interference.

Solution:
We fine-tuned OpenAI's Whisper Small model on domain-specific and noise-augmented data to improve transcription quality in adverse environments.

📄 Read more about ASR →

OCR

The goal is to extract structured text from images of scanned documents. These documents may contain varying fonts, alignments, and noise artifacts that reduce legibility.

Solution:
We fine-tuned the Tesseract OCR engine on our dataset, optimizing preprocessing steps to improve character segmentation and recognition accuracy.

📄 Read more about OCR →

CV

Detect and classify specific target objects within diverse images. Each image may contain multiple instances—or none—of the objects in question, requiring precise localization and accurate classification.

Solution:
We employed strong data augmentation techniques and used RT-DETRv2-M (a real-time detection transformer) for fast and accurate object detection and classification.

📄 Read more about CV →

Get started

Here's a quick overview of the initial setup instructions. You can find a more detailed tutorial, including advanced usage for power users, in the Wiki.

Use this repository as a template to create your own, and clone it into your Vertex AI workbench. You'll want to keep your repository private, so you'll need to create a GitHub Personal Access Token.

You'll also need to initialize the Git submodules:

git submodule update --init

This repository requires Python 3.12 or newer to work. Use conda to create a virtual environment with a Python version of at least 3.12.

conda create --name til python=3.12
conda activate til

Finally, install the development dependencies into your newly created virtual environment.

pip install -r requirements-dev.txt

Understanding this repo

There's a subdirectory for each challenge: asr/, cv/, ocr/, and rl/. Each contains:

• A src/ directory, where your code lives.
  • *_manager.py, which manages your model. This is where your inference and computation takes place.
  • *_server.py, which runs a local web server that talks to the rest of the competition infrastructure.
• Dockerfile, which is used to build your Docker image for each model.
• requirements.txt, which lists the dependencies you need to have bundled into your Docker image.
• README.md, which contains specifications for the format of each challenge.

You'll also find a final subdirectory, test/. This contains tools to test and score your model locally.

There are also two Git submodules, til-25-finals and til-25-environment. finals contains code that will be pulled into your repo for Semifinals and Finals. environment contains the til_environment package, which will help you train and test your RL model, and is installed by pip during setup. Don't delete or modify the contents of til-25-finals/, til-25-environment/, or .gitmodules.

Build, test, and submit

Submitting your model for evaluation is simple: just build your Docker image, test it, and submit. You can find a more detailed tutorial, including advanced usage for power users, in the Wiki.

You'll first want to cd into the directory you want to build. Then, build the image using Docker, following the naming scheme below. You should then run and test your model before using til submit to submit your image for evaluation.

# cd into the directory. For example, `cd ./asr/`
cd CHALLENGE

# Build your image. Remember the . at the end.
docker build -t TEAM_ID-CHALLENGE:TAG .

# Optionally, you can run your model locally to test it.
docker run -p PORT --gpus all -d IMAGE_NAME:TAG
python test/test_CHALLENGE.py

# Push it for submission
til submit TEAM_ID-CHALLENGE:TAG