Inspiration

As students exploring AI, we found it difficult to experiment with NLP + RL algorithms on real-world datasets. Setting up environments, comparing models like PPO, DPO, and GRPO, and understanding how time series integrates with RL required too much time and expertise. We wanted to build a unified, student-friendly platform that makes advanced AI experimentation accessible.

What it does

ThinkLab AI is a web-based learning and research platform that allows users to: Train and compare NLP + RL models on time series and other datasets. Monitor jobs in real time with status updates, logs, and results. Get AI-powered insights and recommendations for hyperparameters, GPU usage, and optimization. Learn through interactive experimentation with visual dashboards and guided assistance.

How we built it

Frontend: Streamlit for interface, Plotly for interactive charts and visualizations. Backend: PyTorch for ML models, custom Transformers for time series, and Gymnasium for environments. RL Algorithms: Implemented PPO, DPO, GRPO, and DAPO for comparison. AI Assistant: Built with Supermemory AI for contextual feedback and guidance. Data Integration: Connected external APIs to fetch real-world datasets (e.g., stock market data).

Challenges we ran into

Standardizing RL pipelines across different datasets. Efficient resource allocation (GPU/CPU/Memory). Designing meaningful comparison metrics and visualizations.

Accomplishments that we're proud of

Built a unified platform where students can experiment with NLP + RL side by side. Designed an AI insights module that provides real-time recommendations. Created an interface that’s both interactive and research-grade. Successfully integrated multiple RL algorithms into one environment.

What we learned

How to merge time series forecasting with RL policies. Building job monitoring systems with real-time feedback loops. The importance of making advanced AI concepts approachable for learners. Deepened technical skills in PyTorch, Transformers, RL algorithms, and visualization tools.

What's next for ThinkLab AI

Expanding dataset support for text, finance, healthcare, and beyond. Adding a Dataset Explorer for uploading and preprocessing data directly. Building a Comparison Dashboard for side-by-side model evaluations. Offering export features (PDF/CSV reports) for research and classroom use. Extending the AI assistant into a full interactive tutor that explains results and concepts.