The Future of Research Is Here! - Chat with *ALL* your data like never before

AI tools are moving research from “search and skim” to “upload and interrogate,” letting researchers ask questions across many PDFs and other documents as if they had a dedicated second brain. The core promise is simple: build an AI knowledge base from a set of files, then use one conversational prompt to extract summaries, next steps, and literature gaps—while grounding answers in the uploaded sources rather than generic training data.

The first tool highlighted, power drill.ai (in beta), focuses on bridging an individual’s data with AI through a code-free workflow. Users create an AI knowledge base by uploading multiple papers into a dataset, then switch the chat mode from a generic ChatGPT-style interface to “Andy research,” which targets the user’s specific database. In a demonstration, the user asks for “obvious next steps” and receives a research-oriented suggestion tied to the uploaded material—specifically pointing toward further investigation into the final location and distribution of SDS in a device. The assistant also provides a more detailed “summary of my research,” along with references pulled from the documents, aiming to reduce the risk of missing key points when dealing with large file collections. The tradeoff is that the workflow is still early: data sources are added one at a time (at least in the current beta flow), and the creator flags that there’s a more robust approach.

That “better way” comes next with quiver, described as a second brain for storing and retrieving unstructured information. Quiver is positioned as more flexible and more controllable than the simpler beta approach: it’s open source, emphasizes secure data handling and user control, and can ingest a wide range of content types beyond PDFs—Word documents, text, URL files, images, and code snippets. The online experience works by uploading files, then chatting with a “default brain” using academic-style prompts. In the example, the assistant identifies a literature gap: missing actual values for collection parameters and resolution, even though those items are mentioned in the dataset.

For maximum control, the transcript recommends running quiver locally rather than using the hosted web app. That setup is more technical: it involves installing Docker, using Super Bass for chunking and storing the uploaded knowledge, and following GitHub “getting started” steps. The payoff is a local interface running on localhost (port 3000), where the brain is configured with an OpenAI key, a chosen model, and a “personality” such as acting like an academic researcher. When asked for a gap in the literature, the local brain returns actionable directions—suggesting areas like new materials, improved fabrication techniques, interface studies, device integration, and durability studies.

Finally, the transcript frames the broader shift: researchers can create multiple “brains” segmented by topic, each with its own uploaded dataset and prompt style. That modular setup is presented as a practical way to keep different research threads organized while querying them with tailored academic guidance. The tools are pitched as early but powerful steps toward a future where research assistants work directly from a user’s own document library.