vishalsingh2972/research-saathi

Research Saathi is a research assistant that splits the work of researching a topic across four separate AI agents instead of asking one model to do everything at once. When someone types in a topic, a Search Agent looks for recent and reliable sources on the web. A Reader Agent then goes through those sources and pulls out the useful information while dropping the rest. A Writer Agent takes that extracted knowledge and turns it into a structured report with sections like an introduction, current trends, key technologies, applications, challenges, and a conclusion. Finally, a Critic Agent reviews that report, checking for gaps, unclear writing, and missing information, and gives feedback alongside the finished report. The whole process runs through an interactive Streamlit web interface, so a user can watch each agent finish its step in real time rather than waiting for one long response. The agents pass information between each other using a shared state that moves through the pipeline in order: search, then read, then write, then critique. It is built with LangChain and connects to Mistral AI's models to power the agents, with an optional integration for LangSmith tracing and the Tavily search API. The project is Python based and uses either pip or the uv tool to install its dependencies. Setup involves creating a virtual environment, installing the requirements, adding a Mistral API key to an environment file, and running the app with a single Streamlit command. The author frames this as a demonstration of why splitting a task across specialized agents can work better than one large prompt, pointing to easier debugging, clearer reasoning, and components that could be reused elsewhere. Planned future additions include exporting reports as PDFs, automatic citations, running agents in parallel, and letting a person review or steer the research mid process. The README's license section is unfinished, so the terms are not fully stated.