selimozten/orin
RL gym for training LLMs on financial text reasoning
What's novel
RL gym for training LLMs on financial text reasoning
Code Analysis
20 files read · 5 roundsA Gymnasium-based RL environment suite for training agents to reason about financial text (earnings reports, news, filings) with synthetic data generation and composite reward systems combining directional accuracy and confidence calibration.
Strengths
Excellent modularity with clear separation between environments, data handling, and rewards; comprehensive test coverage including edge cases like flat predictions and overconfidence penalties; well-documented code with intuitive naming conventions.
Weaknesses
Relies heavily on synthetic data generation which may not fully capture real-world financial text complexity; some reward logic could benefit from additional unit tests for extreme edge cases.
Score Breakdown
Signal breakdown
Innovation
Craft
Traction
Scope
Evidence
Commits
62
Contributors
1
Files
72
Active weeks
1
Repository
Language
Python
Stars
1
Forks
0
License
MIT