@Shannon Rankin has joined the channel
Helloooo I'm in the waiting room (I think?)
I think anyone who joined with the same email address I invited was automatically accepted into the meeting, but I had to manually let folks in who joined with a different email address, and I wasn't watching the waiting room. Sorry!
Ok, so Jules did share what looked like a very reasonable solution to my question about adding "consecutive days labeling photos" to a reward system on our public website.
But, I got stuck on my end trying to get the branch running locally - so I'm taking a break and will try again later.
fixed the game up (well.. you know.. ai) and dropped my chat log for the <5 minute vibe-coded game here: https://github.com/morescode-pm/elephriend-zone/blob/main/development-chat-log.md
@Christine Muthee has joined the channel
had a great time with cursor today building an app to get species lists from the gbif api based on a hand drawn polygon geofence!
it started out as a Flask app that ran great locally, and then I asked it to refactor for front-end only deployment on github pages.
incredible - here's the app! let me know if it works for you
https://morescode-pm.github.io/geofence-polygon/
Amazing! I searched for observations near where I live; polygon drawing was smooth and the results look reasonable.
I think it put some limits on the number of returns or some type of "must be validated" because there are definitely some missing. But still - mostly done through chat messages.
Planning to start using project requirements markdown files next time around.
Jules is very, very good. Insanely good.
One tab per task I asked for the following:
1- a script to process speciesnet output files into a dictionary for our mongodb server (several guiding concepts about the structure and information we wanted at this stage)
Full repo and readme created here: https://github.com/morescode-pm/mongo-ur-utils/tree/main
2- updates to the geofence app above for adding icons of the species found
3- updates to the geofence app to allow for SQL query submission based on the user's gbif account
I'm nervous about this one - to submit a request you need to type in your username and password to the webapp - there's no database where this could be stored, but I'm not sure if the request is being handled with the appropriate privacy considerations.. so much so that I'll probably re-release that option as a "go to gbif, sign up, and submit your query manually by copying and pasting the text below"
Now these took a lot more guidance than before and I had a lot of knowledge about what I wanted done. There was even some small testing I had to do on the side - like format the SQL statement and try querying in google big query vs the gbif online platform with some trial and error.
I started out in vscode with copilot but ran out of monthly requests pretty fast.
What I like about jules is the VERY asynchronous nature - it already assumes that the full repo context is valuable, where vs code seemed to 'pick' files to use for context, sometimes missing a few important ones.
Again, picking my jaw off the floor just to drop it back on the floor.
Very cool! My own biases aside, I agree that Jules occupies a unique place in the space right now: other tools are trying to be more like a pair programmer, Jules is trying to be an excellent developer who is 12 time zones away, which encourages you to interact with it in a different way. Both types of tool are great for different types of task.
Wow.. Ok.. I asked Jules to add a bounding boxes feature to our labeling site and it completely knocked it out of the park in < 4 messages.
"""
In our identify page we have an ecosystem for labeling images - we want to add a bounding box feature so people have to draw a box around what they see in the image - and then when submitting the observation, we need to submit the bounding box coordinates as well.
Please check the full repo & make sure to update models, features, and routes as necessary
"""
"""
ok, this is looking great -
let's make sure the bbox attributes match the schema at: https://camtrap-dp.tdwg.org/data/
Also, there might be more than one species in the photo - so instead of generally drawing a bounding box, let's make it so each species picked has a "draw box" icon and tooltip next to it, so the person identifying the species can draw one or more boxes around each animal - if they select 1 animal, one box must be drawn - if they select 2 animals of a species then 2 boxes can be drawn, and all combinations of that idea.
the icon for draw boxes should indicate what color box outline is used for each species
"""
Then Jules had a few follow up questions and in a short back and forth with some bug fixing (resolution changes were messing up the boxes) - we nailed it down.
The only 'bug' I've been dealing with is sometimes Jules forgets to commit the changes to a branch - which it is more than happy to do when I remind it again.
ok, another day & some more experiences with 🌠 vibes 🤩
1 - This was big. Our data is in MongoDB to run the website - but our eventual goal that nobody really wanted to take on is : get the data into CamtrapDP format.
- I downloaded example datasets and schema from camtrap dp into a repo
- I asked chatGPT to structure a script for me to download an example from our MongoDB
- I casually asked Jules to take a look at my target schema and current schema and build a tool to process the files.
looking at the database files in the mongo_exports folder saved as samples and schema json files -- please write some python functions to convert what you find to camtrap dp format.
camtrap dp format is defined in camtrap_standards/definitions
there are example csv files in camtrap_standards/examples
The goal is to be able to download all of the data in the mongo database and format it into csv files with the information that camtrap dp specifies.
try to get as much data as possible, but if there are missing fields, it's ok to leave them empty.
IT WORKED. It included a lot of reformatting and edge cases and even built a testing dataset. Months and months of time saved - incredible.
2. I've noticed that Lovable is very very reliant on creating apps only in their structure - so much so that asking it to "do nothing and wait for my next message" creates a TON of files. And, it definitely ignores instructions that ask it to try another infrastructure.
I asked it to make me a 'conservation dashboard for analysis' and it .. looks the same as everything else Lovable has made me and is essentially vaporware (looks nice, requires a lot more input) - maybe I'm using the free version too creatively?
3. Testing Replit right now - trying to have it help start up a similar dashboard using streamlit. This one actually listened - this is my first time testing replit and I'll report in after we see. So far so good
same prompt for both:
I want to create a dashboard using streamlit where I can visualize data from my wildlife cameratraps stored in camtrap dp files "media.csv" "deployments.csv" and "observations.csv"
using information in my observations.csv
the tool should be able to pick the species scientific name (using common names) from a drop down -
also pick at the higher level the "observationType" - that influences the species scientific name from the drop down.
I want to be able to overlay these things using check boxes.
Generate some interesting visuals based on conservation biologist research paths. For example, circadian rhythms, light sources, abundance, counts etc.
be sure to not double count items where the event id is the same in observations.csv
you can use the information about geolocation from the media exifData or linked deployments.csv as well.
the dashboard should be rich and include example images from the web or from the publicURLs in media.csv
Amazing! Particularly since it seems unlikely that it had lots of information about Camtrap DP in training. Hard to say whether it just pored over the data in your .csv files and inferred the format, or read the Camtrap DP specs.
Totally amazing. No way to know for sure - but based on how I've seen Jules digest giant codebases before I'm inclined to believe it actually read the .csv and .json files.
Reviewed the code and what it wrote is legit!
verdict on replit - free trial exausted quickly - it did a really decent job.
A bit of a bummer that the majority of my free credits were consumed with error fixing - but in the end it made some interesting visuals (and some... bad visuals) based on some choices it made without guidance.
was able to download all the files in a .zip and will use it as a starting point
@Eric Cunningham has joined the channel
