WikiTrends

Another small devlog, but a good feature.
I added a “Long Version” page that lists more articles and has a longer analysis, so you can look at whichever version you want, based on your attention span or your available time.

A quick devlog.
I fixed the demo, meaning it shows the features when used and not daily, as that would only be boring and cause problems.
Added a greet message for the new users who subscribe for the telegram notification and fixed general details.

Huge devlog this time hehehehe
The entire thing is hosted 24/7 and changes at 8am UTC (if i’m not wrong), same time at which the telegram bot updates the subscribed users.

I added:

• nice dynamic frontend hosted on a web server
• long version and short summary of the analysis (the analysis changes daily)
• telegram bot sharing the latest analysis daily (while fighting with rate limits)
• bot users database (while learnig sqlite)
• finally wrote a readme

Fixed A LOT of things be it typos, errors or any small thing.
I think i coded much more than this but it took so long I forgot
Took me a loooong untracked time to make my demo work on a VPS, Oracle’s fault…

To-do:

• test the bot with more than 30 subscribed users
• possibly add a graph for the trends
• make a mobile app or any alternative notification system
• website notifications
• custom telegram message time

Have fun! :3

THIS DEVLOG CONTAINS A SHOWCASE!!!
Small devlog for a kinda quick job.

TL;DR: i managed to use venv and git correctly cause i mess things up 99% of the time when using those, looks promising.

I set up the github repo, made it work and synced it on both my devices.
That’s great cause i was finally able to test my code on more than one platform and it works perfectly, only dependency that’s not managed by python and its libs is ollama, or any other API you wanna use.

I fixed the requirements.txt i previously made, it was made in a bad venv and full of libs that are not needed anymore, now it’s light and contains only the things that are strictly necessary for this to work, as it should be.

I also made some small tweaks in my code and made sure all the references to “context” were removed or fixed, as it was breaking the code and it wasn’t needed at all.

Github repo is almost complete, telegram bot support is also almost complete, and the project might be close to a beta release suitable for shipping :)

Some good updates this time!

As i wrote in the last devlog, the LLM i was using is not available anymore with the API i was using (Cerebras).
I was using cerebras because that’s what i started learning LLMs in python with, but now i needed something local to make it better.

I now use ollama cloud to communicate with a crazy good deepseek model (v3.1 671b), the results are excellent (better than they were with gpt-oss-120b) and the analysis and predictions actually make sense now. I fixed some minor things too like memory, timestamps and output.

I also fixed some small things in the memory section: the date given to the LLM no longer corresponds to the timestamp (the time it was written to memory), so now it’s possible to fetch older data and still have the right time in the prompt.

I removed the context part, it was a useless file containing just a string, i wrote the string in the code itself.

I removed a function that listed the models currently available on cerebras as i no longer use cerebras.
Also added some functions that save the prompt to a file so i know what’s been sent to the LLM, write the output to a markdown file and convert them to a pdf so that the latest analysis can be easily read.

My code is still kinda messy and incomplete, it’s full of functions and still missing the user integration part.

To-do:

• fetch custom date without changing the url
• use external config file for API url, key etc.
• telegram bot/web app or client part
• graphs showing what’s been trending(?)
• option to check daily, weekly, etc.

Thanks for reading, hope you like what I’m doing.

NEW DEVLOG!!! I just came back from an event about start ups and coding, placed third :)

I finally fixed the way memory works, now it will store all the data the Wikimedia API returns, with the corresponding analysis and predictions, so that the new analysis can be based on the latest one, but also on the last couple entries, for better accuracy.

Both API responses and memory are saved to data.json and memory.json, if deleted, data will be recreated next time the function is called, if memory is deleted, well, the analysis won’t take count of the old ones. And it will be recreated too.

The LLM provider i’m relying on is pretty unstable, one day they allow you to use GPT, one day only llama and so on, that’s not really nice.

That’s kind of a big problem since i was using GPT-oss which is not available anymore (for now). I coded this part testing it offline, as it is just a better version of the one in my first devlog, so there was nothing API or AI related to test.

I’ll use a local LLM soon, even though i don’t have a dedicated machine to run it on. I may as well create some categories directly in my code and use those for the analysis, but it won’t be very accurate and it will be way worse than it is now.

FIRST DEVLOG NEW PROJECT!! :D
Nice idea i had thanks to my bro, had to try some “data analysis” sooner or later, so here it is.

Added:
-first attempt to use an LLM via API, not the best but it works
-function to fetch the trends data from wikimedia’s API
-another function to filter the data in order not to show the main page and search page as trending
-made a context instruction for the LLM, it won’t reply if the prompt or data is out of context
-made a memory instruction for the LLM, not used yet, but it’s there for future in-depth analysis based on data that was already analyzed
-collected data is exported to json file
-final output includes top 5 trending articles, general overview and predictions for the next time period trends
-output is also saved in a MD file
-split the code in functions simply because it doesn’t have to be in a loop, it will be called only after certain events which will be added soon