3. Data, Science & AI

Validates trip data quality using four algorithms: distance sum, fuel consumption, efficiency range, and deviation.

Reconstructs trip paths by filling gaps between checkpoints using a hybrid GPS (70%) and address (30%) template matching algorithm with high confidence.

Validates CSV files for data quality, consistency, and completeness by checking missing values and format issues.

Manages GPU rentals, ML training (PyTorch/TensorFlow), and compute resources on Basilica's decentralized marketplace via CLI workflows.

Automates end-to-end validation of credit risk models in banking, generating reports with key metrics and regulatory compliance for Kazakhstan.

Enables seamless integration of multiple LLMs (Claude, GPT, Gemini, Ollama) with API management, prompt engineering, and multi-model orchestration.

Enables application of offline RL algorithms (CQL, IQL, BCQ) using fixed datasets without environment interaction.

Extracts text, fills forms, and analyzes document structure from PDF files for data processing and analysis.

Enables implementation and mastery of policy gradient reinforcement learning algorithms including REINFORCE, PPO, and TRPO for direct policy optimization.

CLI tool for generating natural-sounding speech from text using OpenAI's TTS API.

Designs and validates reward functions for reinforcement learning using potential-based shaping techniques.

Provides utilities for creating and managing custom reinforcement learning environments with Gymnasium API, including spaces, wrappers, and vectorization.

Provides implementation and explanation of value-based reinforcement learning algorithms including DQN, Double DQN, Dueling DQN, and Rainbow for discrete action spaces.

Enables implementation and mastery of model-based reinforcement learning algorithms including Dyna-Q, MBPO, and Dreamer, utilizing world models for planning.

Provides foundational knowledge of Reinforcement Learning theory including MDPs, Bellman equations, and value iteration methods.

Enables implementation and mastery of actor-critic reinforcement learning methods (A2C, SAC, TD3) for continuous control tasks.

Guides implementation of reinforcement learning exploration strategies including ε-greedy, UCB, curiosity-driven, RND, and intrinsic motivation.

Routes users to appropriate deep-RL skills based on problem type and algorithm family for efficient AI model selection.

Queries real-time parking availability in Taipei, showing status, location, contact, and payment details for all vehicle types.

Provides rigorous evaluation protocols for reinforcement learning models, including statistical methods, train/test discipline, and generalization metrics.

Manages ML experiments including execution, parameter sweeps, result comparison, and log maintenance via CLI.

Systematic debugging tool for reinforcement learning models, diagnosing learning failures and common bugs through structured diagnosis trees.

Enables implementation and optimization of multi-agent RL algorithms including QMIX, MADDPG, and CTDE for coordinated agent decision-making.

Converts text to natural-sounding speech using the local neural TTS system Piper with multiple voice options.