ahmetk1212/sciscout

SciScout is an AI research agent that answers scientific and engineering questions by searching trusted academic and technical sources, reading the papers it finds, and writing back a synthesized answer with citations attached. It can take a question in Turkish or another language and translates it into an English academic search query before running the search, so people who do not normally write in English can still use it. Under the hood it works as a small pipeline of three roles. A researcher step translates the question and searches across multiple sources, ranking what it finds. A reader step pulls out the actual content, including reading full PDF text when just the abstract is not enough. A synthesizer step then writes the final answer with inline citation footnotes pointing back to the sources used. It searches twelve sources in total, including arXiv, Semantic Scholar, PubMed, Europe PMC, NASA ADS, DOAJ, Wikipedia, GitHub, StackExchange, Google Scholar, Reddit, and a general trusted-domain web search provider. For the language model itself, it routes between several providers such as Groq, Cerebras, and Gemini, automatically falling back to another one if the first is unavailable. You can use it from a command line interface with commands to ask a question, search a single source directly without invoking the model, run an interactive chat session, or start it as a FastAPI web server with health, ask, streaming, and search endpoints. Getting it running means cloning the project, creating a Python virtual environment, installing it, and filling in a few API keys in an environment file, several of which have free tiers. The README lists known limitations: some of the sources it queries can rate-limit or throttle requests, and its Google Scholar search relies on scraping the site, which can break if Google changes its page layout. It is released under the MIT license.