What is Scienvera?
I am excited to share my new project – a Software as a Service (SaaS) platform that changes how users find and access research articles. This platform uses advanced artificial intelligence to search the large databases of PubMed and EUPMC, helping you easily identify the most relevant and reliable articles for your research needs.
What was my motivation?
Well, I am actually studying for genetics and bio engineering, so I regularly need to check every single argument that I state in academic papers, laboratory reports, review papers, or even simple presentations. But medical field is huge.
Like actually huge. Most of the time, you need to search something so basic that you can’t find it on top 10, 20 results. That is because there are so many specific articles that the title matches with your text, but only to realize that you probably can’t use that article when you actually read the abstract.
On top of that, once you find your perfect article that was on literally 2nd/3rd page on your google search, you realise that you spent so much time just to find related ones.
What if there was a tool that automatically searches on your behalf, and actually cross checks the abstract?
My Structure & Tech Stack
Well, the story that i am telling later will have much more detail about my changes, but this is my final structure and everything. Well i dont feel the need to specify their languages, they are pretty self explanatory.
Front-End:
- Used React for framework for simplicity.
- Using Kinde authentication provider for more streamlined security
- Mantine UI Kit for rapid designing. (I am so bad at designing please help)
Back-End (This is the good one):
- Kubernetes for cluster orchestration
- Docker for containerization
- Gitlab for automated CI/CD pipelines
- Containerized nginx for react side
- Spring Cloud Gateway for routing and load balancing for kubernetes
- Artemis JMS for message queues for load balancing between modules
- PostgreSQL and MySQL for vector management and user database respectively
- Spring AI for tokenization, embedding and similarity checking
- Spring Security for OAuth2 authentication with Kinde
- Cloudflare for network filtering/protection
- Stripe for purchase integration
- Spring JPA for Hibernate
- Spring Web for remaining functionality
Huge problem with AI Tools
I mean there are not many tools that are ACTUALLY HELPS ACADEMICALLY. Well yeah there are bunch of ChatGPT wrappers that can read pptx, pdf and answer, or turn your slideshows into 5 sentence garbage summary, or listen to your real life lessons and creates 2 page pdf file(btw who regulates that audio data, thats sensitive asf???), but actually cross check? Yes there are some chat bots that have “capability to search internet” and can find sources for your text, but for my experience, it has same problem.
LLMs give text based on their huge knowledge base whether it searches the web or not. Sure, they increase the chance of selecting relating embedded vector, but they do NOT incorporate the search results into their own network. They are just adding it as context. So, even if they give sources, chances are they just have that data on their back already, and give some google results that can potentially match using their title. Even if they go read the abstract, they are only searching through google/non academic search engines. While my tool can search up to limitless related articles, and only articles.
One more issue is knowledge date. Science is fast-paced by their nature. And LLM models does not include newest academic releases. If you search broad and general things, sure go ahead. But when i try to search 2024 article with 2020 model LLM, even if it searches the net and finds the 2024 article, it will not read the whole article, and even if it can read, it maybe can not connect the bridge due to date, so it can spit non sense.
But academic research is not based on article titles. Rather, it is based on much more complicated abstracts, full texts and conclusions. That is why I created this.
My Writeup
If you want to read my process, you can look up part 1 and part 2. If you want to visit the website, here it is.