I have just finished writing my MSc Thesis and I think that I made a decent literature review in a fairly reasonable time. So I am writing this post to share a bit how to use different tools to have a very fluid writing process. I’m no prompt engineer or whatever the current buzzword is. I am just using some tools for my benefit as things should be I guess.

Tool list

• Zotero
• Zotero Connector browser extension
• BetterBibTex for Zotero
• Your University VPN
• Some LLM tool, preferrably some pro version or self-hosted
• Latex/Overleaf
• ‘Scopus AI’ or ‘Web of Science Research Assistant’. I do not endorse any in particular. Maybe your university gives you access to another one of these tools, take a look at mine.

Setup

Configure and install betterbibtex, it’s really useful. Configure only the fields you need such as authors, title, year and journal. Create a new collection on Zotero (if you don’t already have one).

My pipeline

I will start with a hypothetical example, and I will write latex in code to specify I am talking about.

Write some phrases you know are true because of some tribal knowledge you know from your experience in research.

An important step in the development of neural networks was the creation of skip connections:
they allow for a direct flow of information into downstream layers, which allows the network
to combine higher level features to lower level information.
Skip connections allowed the development of Residual networks and U-Nets, as well as different cutting edge models.

Now copy and paste into some LLM with a prompt like this:

Help me write the literature review section of my dissertation.
I want you to insert (CITE: some reference) in parts of the text that require citation.
Output only latex:

An important step in the development of neural networks was the creation of skip connections:
they allow for a direct flow of information into downstream layers, which allows the network
to combine higher level features to lower level information.
Skip connections allowed the development of Residual networks and U-Nets, as well as different cutting edge models.

Then, the type of response that you might get:

An important step in the development of neural networks was the creation of skip connections
(CITE: ResNet paper, He et al. 2016): they allow for a direct flow of information into downstream layers,
enabling the network to combine high-level features with lower-level information
(CITE: paper explaining skip connections and gradient flow). Skip connections have been fundamental for
the development of Residual Networks (ResNets) (CITE: He et al. 2016 ResNet),
U-Net architectures (CITE: Ronneberger et al. 2015 U-Net), and numerous state-of-the-art
models (CITE: review paper on modern CNN architectures or DenseNet paper).

Now up for some work! The LLM did most of the heavy lifting, now we just have to actually search for some references. ResNet paper is easy, you just have to search on your favorite search engine, U-Net follows the same idea, so let’s focus on using some cool tools. Now for the 2 remaining references, we can use some pretty convenient tool like I mentioned before. My university gives me access both to Scopus AI and WoS Research Assistant, so let’s use the first one.

In your favorite tool, let’s start by writing for example “paper explaining skip connections and gradient flow” and see what type of results we get:

Sweet! It very quickly made a small literature review (based on the abstracts). Let’s start opening the second reference as it has a promising title:

After reading a bit, it seems that their literature review and claims really support what we are stating, so that passes! ✅ (no pun intended). When you have the publication you want to add, click on the Zotero extension, it will automatically add the publication to your collection:

Awesome, right? Now zotero will automatically highlight it in the collection, and if your plugin is correctly configured, it will show you the “citation key” field:

Copy that, and now in your latex you preemptively paste the key inside the paste command inside a latex \cite{} command:

An important step in the development of neural networks was the creation of skip connections
(CITE: ResNet paper, He et al. 2016): they allow for a direct flow of information into downstream layers,
enabling the network to combine high-level features with lower-level information
\cite{shenNovelConnectivityDeep2017}. Skip connections have been fundamental for the development of
Residual Networks (ResNets) (CITE: He et al. 2016 ResNet),
U-Net architectures (CITE: Ronneberger et al. 2015 U-Net),
and numerous state-of-the-art models (CITE: review paper on modern CNN architectures or DenseNet paper).

Now you do the same idea for the other references as well. Previously I said preemptively, because now comes the most satisfying part: when you are done, do the final trick: select all entries in Zotero, and click my favorite “Copy BibLaTex to clipboard”:

Now this is my favorite part because, if you happen to use overleaf like me, it’s kind of a pain in the ass having to go back to your local filesystem and reuploading your references. So you just open your references.bib or whatever your file is, and paste your freshly baked references 🍞 into your document.

Final words

This is basically how I did my literature review, I went from 4 to 84 literature references in probably less than 2 weeks without having to do any all nighter (in fact working quite alright hours on this work). Although I have shown a very simple example, it can be easily extended to more complex cases as well. One shortcoming I see is when the authors do not put everything into the abstract (which I guess is the main source of info for those AI assistants) and therefore can at best infer what you want it to support. Another thing I just thought is that this is more useful for surface level literature review, mostly by giving more volume towards what the panorama of your research topic is tilted towards, than to give some deep analysis (quite obviously), which is another problem on its own. But I guess that’s nothing to do about it, at least we are not yet referencing some tiktok videos or skibidi toilet videos (this is a pretty good text btw).

If you liked this guide, thank you for your attention and visit me regularly 😄. If you didn’t, it’s your problem because this website is mine and remember that any human generated information in this dead internet is a gift and an oasis of life, so rethink your attitude 😡.