3. Data, Science & AI

Reviews mathematical documents and proofs for errors, generating detailed correction reports with severity classification and saving to a dedicated review folder.

Builds and operates audiobook TTS pipelines with XTTS training, WhisperX alignment, voice coloring, and persona inference capabilities.

Enables file-based messaging between AI agents for cross-project communication, including inbox checks, bug requests, and issue acknowledgment.

Converts scanned PDFs and images to structured Markdown using Mistral AI's OCR capabilities for efficient document processing.

Provides expert guidance on database schema design and SQL query optimization for performance and scalability.

Generates PDF test fixture images using HuggingFace FLUX.1-schnell AI, supporting mermaid diagrams and placeholders.

Generates publication-quality figures from code analysis using multiple backends (Graphviz, Matplotlib, Plotly) for technical domains like mathematics and aerospace.

Generates new AI agent capabilities through user-defined skill creation, expanding the agent's functional scope.

Provides PostgreSQL-based memory storage for AI agents to store and retrieve decisions, learnings, and observations efficiently.

Extracts plain text from PDF documents using the pdftotext command-line tool for efficient data processing.

Guides development of AI systems for scientific discovery, covering design, validation, release strategies, and evaluation for real-world impact in scientific domains.

Analyzes image quality evaluations per page by directly querying the database for accurate story-based results.

Provides insights on spatial intelligence in AI, covering 3D modeling, computer vision evolution, and AI research strategy based on Fei-Fei Li's work.

Enables live web data extraction, URL mapping, and page crawling for data acquisition when built-in tools are unavailable.

Reference guide for Physical Intelligence's robotics foundation models, covering pre-training/post-training paradigms, pi-zero architecture, and generalization techniques.

Framework for building scalable web crawlers and extracting structured data from websites efficiently.

Builds LLM applications with LangChain, enabling chains, agents, memory systems, and tool integrations for conversational AI and document question-answering.

Builds type-safe AI agents and LLM applications using PydanticAI with structured outputs and dependency injection for production Python AI development.

Converts PDF, Word, Excel, and other files to Markdown for AI analysis, enabling content extraction and data preparation for language models.

Extracts text and data from PDF documents, including OCR for scanned files, supporting single and batch processing.

Automates data extraction and processing from diverse documents using Snowflake Cortex AI, supporting invoices, contracts, and reports.

Initializes AI Engineering Harness by creating AGENTS.md and thoughts/ directory structure for context engineering in new repositories.

Enables local inference, embedding generation, and fine-tuning of transformer models using Hugging Face Transformers library.

Executes Phase 2 plan validation concurrently using Cursor and Gemini AI models to accelerate verification.