The noob’s corner
Some background and context
I tried deepset.ai’s deepset studio as part of a requirement for my internship at Quest1, where I was asked to work with it and explore the features and developer experience. As someone who had never stepped into this side of AI before, I was nervous but definitely curious 😊 .
Taking the first steps
deepset Studio promised to be beginner-friendly and with the free version available, I was all ready to step in and see what I could do with it. First off, the UI was such a breath of fresh air when compared to all the fancy, neon-themed websites and apps these days. It’s clean, simple, and super easy to navigate. Uploading my healthcare-related documents was a piece of cake. I could organize them neatly, and the whole thing felt straightforward and smooth. Honestly, I didn’t feel intimidated at all, which, as a total newbie, was a huge win for me.
Building a simple RAG pipeline
Building the pipeline and implementing it sounded like it’d be complicated, but it wasn’t. Deepset Studio offers step-by-step tutorials that walked me through the entire process and also had already available pipelines to experiment with. Setting up the retriever to fetch relevant passages and integrating the RAG-chat-gpt4o-en model for generating responses was shockingly simple. And here’s the cool part: I could check out the code behind each step. For someone like me, who’s curious about what’s happening under the hood, this is a game-changer. Seeing and understanding the code for these processes is extremely crucial if we need to wrest control for customizations.
Running the pipeline
Once my pipeline was up and running, I got to play around with it in the playground. This was my favorite part. It was interactive, easy to use, and gave instant feedback. The model’s responses were spot on, especially for healthcare-related queries. For instance, it gave super-detailed advice on mold prevention, sun allergies, and pet allergies. The accuracy blew me away.
First impressions
So what actually stood out to me? The whole experience was just so user-friendly. It felt like the platform was designed for people like me, who are new to this. Everything was laid out logically, and I never once felt lost.
The tutorials were a godsend. They made the learning curve almost nonexistent and gave me the confidence to experiment. Code exporting was a feature I didn’t know I needed until I used it. Exporting the code for every process not only satisfied my curiosity but also gave me the option to tweak things if I wanted to. It struck the perfect balance between simplicity and giving users room to grow.
For a free tool, deepset Studio is seriously impressive. It didn’t feel like I was missing out on any essential features, and that’s rare.
Did I feel like there were any downsides? Frankly, there weren’t many. Maybe if you’re looking to do super complex customizations, you’d have to step outside the platform a bit, but for me, it had everything I needed. Adding more advanced tutorials for those wanting to level up would be a nice touch.
deepset Studio has been such a fantastic experience. It’s simple, effective, and perfect for building RAG models, especially if you’re just starting out. It gave me the confidence to dive deeper into this field, and implementing what I needed was seamless. For a free tool, it’s downright impressive. As someone who was new to RAG models and this side of AI, it was great to use.
Sample Images
Figure 1.1 — deepset Studio’s playground UI in action
Figure 1.2 — Sample prompts with responses based on the Healthcare dataset
Figure 1.3 — An example of passage responses with references
Acronyms and Definitions
UI: User Interface
AI: Artificial Intelligence
RAG: Retrieval-Augmented Generation is an AI framework that enhances Large Language Models (LLMs) by connecting them to external knowledge bases, allowing them to access and reference information beyond their training data, leading to more accurate and contextually relevant responses
GPT: Generative Pre-trained Transformer is a type of large language model (LLM) and a prominent framework for generative artificial intelligence. It is an artificial neural network that is used in natural language processing by machines. It is based on the transformer deep learning architecture
Pipeline: In RAG, a pipeline refers to the end-to-end workflow that integrates a retrieval system with a generation model to enhance the quality and relevance of responses.
Query: A query is the input question or search request made by the user to the RAG system.
Passage: A passage is a small chunk of text retrieved from a knowledge source (such as a document, Wikipedia, or a database) to provide relevant context for answering the query.
Retriever: A retriever is the component responsible for fetching relevant documents or passages from an external knowledge source based on a given query
References
\[1\] deepset.ai — https://www.deepset.ai/
\[2\] The deepset Studio — https://www.deepset.ai/deepset-studio
\[3\] Sample Healthcare dataset — https://www.mayoclinic.org/diseases-conditions
\[4\] Know more about deepset.ai’s partnership with Quest1
Acknowledgements
Firstly, I would like to thank Quest1 for giving me the opportunity to explore deepset.ai’s deepset Studio for my internship. Their guidance made this learning experience smooth and very insightful. Also, a big thanks to deepset.ai for providing a well-documented and beginner-friendly platform that made working with RAG pipelines intuitive. Lastly, I appreciate the open-source community and platforms like Mayo Clinic for providing reliable datasets that were invaluable in testing my RAG pipeline.
About Tanushree Vijaysanan
I’m a college student specializing in AI and Machine Learning, with experience in ML models, quantum computing, and full-stack development. I love exploring emerging tech and tackling complex problems.
I co-founded Blitz, an outing planning app that uses ML for personalized recommendations. I’ve led IBM quantum lab demos, debugging competitions, and even published a patent on a blockchain-based stock market prediction system.
Skilled in Python ML libraries like Scikit-learn, TensorFlow, and Pandas, I’ve worked on projects like neural networks for pothole detection. I’m also part of ACM-W SIST’s tech team and have anchored events with big names.
Lately, I’ve been diving into RAG models, especially for healthcare AI, always looking for ways to build impactful solutions.