LLM App

Pathway LLM App monitors and syncs documents from heterogeneous data sources (file systems, Google Drive, SharePoint, S3) with automatic change detection and incremental updates. The framework uses Pathway's reactive dataflow engine to detect source changes and propagate them through the pipeline without full re-indexing, enabling live document ingestion at scale across millions of documents while maintaining consistency.

Pathway LLM App includes pluggable document parsers that extract text and structured metadata from multiple formats (PDF, DOCX, TXT, HTML, etc.) while preserving document structure and semantic information. The parsing layer integrates with libraries like PyPDF2 and python-docx, handling format-specific quirks and producing normalized output that feeds into the embedding and retrieval pipeline.

Pathway LLM App includes Multimodal RAG capabilities that process both text and images, enabling RAG systems to retrieve and reason over visual content. The framework integrates vision models (GPT-4V, etc.) to understand image content, extract text via OCR, and generate descriptions that are indexed alongside text chunks. This enables unified search over mixed-media documents.

Pathway LLM App provides document indexing capabilities that create searchable indices over document chunks using both vector embeddings and keyword matching. The framework supports full-text search with inverted indices, enabling fast keyword-based retrieval alongside semantic vector search. Hybrid search combines both approaches to improve retrieval precision and recall.

Pathway LLM App integrates with LangGraph to enable multi-step reasoning agents that can decompose complex queries into subtasks, retrieve context iteratively, and make decisions based on intermediate results. Agents can use tools (search, calculation, etc.) and maintain state across multiple reasoning steps. This enables more sophisticated query answering than single-step RAG.

Pathway LLM App provides pre-built pipeline templates for specific use cases including Slides AI Search (searching presentation content), Unstructured to SQL (converting unstructured documents to structured data), and Drive Alert (monitoring cloud storage for changes). These templates are ready-to-deploy examples that can be customized for specific domains, reducing development time for common patterns.

Pathway LLM App uses declarative configuration files (app.yaml) to define entire RAG pipelines without code changes. Configuration specifies data sources, document parsing, chunking, embedding models, LLM providers, indexing strategy, and retrieval parameters. This enables non-developers to customize pipelines and developers to manage multiple pipeline variants without code duplication.

Pathway LLM App provides configurable text splitting strategies that divide documents into chunks optimized for embedding and retrieval. The framework supports both fixed-size chunking and semantic-aware splitting that respects document structure (paragraphs, sentences, sections), with configurable overlap to maintain context between chunks. Chunk size and overlap parameters are tunable via the app.yaml configuration system.

Pathway LLM App integrates with embedding models (OpenAI, Mistral, local models) to convert text chunks into vector representations, then indexes these vectors for efficient similarity search. The framework supports both pure vector search and hybrid search (combining vector similarity with keyword matching), with the indexing strategy configurable via app.yaml. Vectors are stored in an in-memory or persistent vector index that supports approximate nearest neighbor queries.

Pathway LLM App processes user queries through a retrieval pipeline that finds relevant document chunks from the indexed corpus. The framework supports query rewriting (reformulating queries for better retrieval), context retrieval (finding top-K similar chunks), and ranking strategies to order results by relevance. Retrieved context is passed to the LLM along with the original query to ground the response in retrieved documents.

Pathway LLM App provides a unified interface to multiple LLM providers (OpenAI, Mistral, local models via Ollama) for generating responses grounded in retrieved context. The framework handles prompt construction, context injection, and response streaming, with provider selection configurable via app.yaml. Responses are generated by passing the user query and retrieved document chunks to the LLM, enabling RAG-based question answering.

Pathway LLM App automatically exposes the RAG pipeline as HTTP REST endpoints that accept queries and return LLM-generated responses with retrieved context. The framework handles request routing, response serialization, and optional streaming of responses to clients. API endpoints are generated from the pipeline configuration without manual endpoint definition, enabling rapid deployment of query interfaces.

Pathway LLM App includes a Streamlit-based user interface that provides an interactive query interface for the RAG pipeline. The UI allows users to submit queries, view generated responses, and inspect retrieved context documents. The Streamlit app is automatically generated from the pipeline configuration, enabling rapid deployment of user-facing interfaces without frontend development.

Pathway LLM App includes an Adaptive RAG pattern that selects retrieval strategies dynamically based on query characteristics. The framework analyzes incoming queries to determine whether to use vector search, keyword search, or hybrid search, optimizing retrieval for different query types without manual configuration. This pattern improves retrieval quality by matching retrieval strategy to query intent.

Pathway LLM App supports Private RAG deployments that use local embedding models (sentence-transformers, etc.) and local LLMs (Ollama, LLaMA, etc.) instead of cloud APIs. This pattern enables RAG applications to run entirely on-premises without sending data to external services, addressing privacy and compliance requirements. Local models are integrated via the same provider abstraction as cloud models, allowing seamless switching.