NextAI
Conversational document AI agent for text- and voice-based PDF queries across multiple documents.
Problem
Working across long PDFs and multiple documents means constant scrolling and searching to find the relevant passage, and there is no fast, hands-free way to ask a question and get a grounded answer.
Solution
NextAI is a conversational document assistant that answers text and voice questions across multiple PDFs and long-form content, returning answers grounded in the source documents.
Technical Approach
A React front end and Express back end coordinate document ingestion, retrieval, and chat.
A custom AI agent built on LangChain and GPT-5 handles multi-document retrieval and long-form content analysis.
In-memory vector caching stores embeddings so repeated queries against the same documents avoid redundant work and reduce latency.
Implementation
- Built the AI-powered PDF chat assistant supporting both voice and text querying using React, Express, and LangChain.
- Engineered a custom AI agent with GPT-5 to facilitate multi-document retrieval and long-form content analysis.
- Optimized document retrieval efficiency with in-memory vector caching to reduce latency.
- Integrated speech-to-text (STT) and text-to-speech (TTS) to improve accessibility and enable hands-free usage.
Key Features
In-memory vector caching
Caching document embeddings in memory keeps repeated retrievals against the same documents fast, avoiding recomputation and lowering query latency.
Voice-first accessibility
Speech-to-text and text-to-speech make the assistant usable hands-free and more accessible, so users can ask and hear answers without reading or typing.