Francisco T. Barbosa
@tuminha.bsky.social
Dentist + digital tech enthusiast. Founder of Periospot.com 🔬 | Exploring the power of AI & ML to build tools that empower people and enhance everyday tasks.
Celebrating 100 Repos on GitHub. 🎊🍾
November 17, 2025 at 9:56 AM
Celebrating 100 Repos on GitHub. 🎊🍾
**Requirements:**
- 2+ years in digital marketing/coordination
- CMS, social media management
- Excellent English
- M365/Teams proficiency
**Bonus:**
- Healthcare/education/NPO experience
- Cvent/CRM/SEO skills
- EU languages
👉 Apply: careers.envistaco.com/job/2261483...
- 2+ years in digital marketing/coordination
- CMS, social media management
- Excellent English
- M365/Teams proficiency
**Bonus:**
- Healthcare/education/NPO experience
- Cvent/CRM/SEO skills
- EU languages
👉 Apply: careers.envistaco.com/job/2261483...
Digital Marketing & Council Coordinator - Job Details
Envista has an opening for a Digital Marketing & Council Coordinator in Kloten, Zurich, CH
careers.envistaco.com
November 11, 2025 at 12:20 PM
**Requirements:**
- 2+ years in digital marketing/coordination
- CMS, social media management
- Excellent English
- M365/Teams proficiency
**Bonus:**
- Healthcare/education/NPO experience
- Cvent/CRM/SEO skills
- EU languages
👉 Apply: careers.envistaco.com/job/2261483...
- 2+ years in digital marketing/coordination
- CMS, social media management
- Excellent English
- M365/Teams proficiency
**Bonus:**
- Healthcare/education/NPO experience
- Cvent/CRM/SEO skills
- EU languages
👉 Apply: careers.envistaco.com/job/2261483...
Join us for a role combining digital marketing and scientific impact!
**Digital Marketing & Council Coordinator (Kloten, Zurich area)**
Be part of our expert team, managing web/SEO/GA4, social media (Sprout), council coordination, and webinars.
**Digital Marketing & Council Coordinator (Kloten, Zurich area)**
Be part of our expert team, managing web/SEO/GA4, social media (Sprout), council coordination, and webinars.
November 11, 2025 at 12:20 PM
Join us for a role combining digital marketing and scientific impact!
**Digital Marketing & Council Coordinator (Kloten, Zurich area)**
Be part of our expert team, managing web/SEO/GA4, social media (Sprout), council coordination, and webinars.
**Digital Marketing & Council Coordinator (Kloten, Zurich area)**
Be part of our expert team, managing web/SEO/GA4, social media (Sprout), council coordination, and webinars.
Full analysis with code and methodology: github.com/Tuminha/Per...
And full story here: substack.com/home/post/p...
Built this for my @Periospot dental education platform but the lessons apply everywhere.
And full story here: substack.com/home/post/p...
Built this for my @Periospot dental education platform but the lessons apply everywhere.
The Perfect Crime of Viral Content: What Ilya Sutskever’s Detective Tale Taught Me About Social Media
3,745 posts, temporal validation, and the rigged game of reach behind engagement.
tuminha.substack.com
November 3, 2025 at 3:00 AM
Full analysis with code and methodology: github.com/Tuminha/Per...
And full story here: substack.com/home/post/p...
Built this for my @Periospot dental education platform but the lessons apply everywhere.
And full story here: substack.com/home/post/p...
Built this for my @Periospot dental education platform but the lessons apply everywhere.
Final thought: I spent months trying to engineer features to predict viral posts.
The model taught me I was asking the wrong question.
Not "what makes posts viral?" but "why do we overestimate content's impact?"
Distribution > Creation. Every time.
The model taught me I was asking the wrong question.
Not "what makes posts viral?" but "why do we overestimate content's impact?"
Distribution > Creation. Every time.
November 3, 2025 at 3:00 AM
Final thought: I spent months trying to engineer features to predict viral posts.
The model taught me I was asking the wrong question.
Not "what makes posts viral?" but "why do we overestimate content's impact?"
Distribution > Creation. Every time.
The model taught me I was asking the wrong question.
Not "what makes posts viral?" but "why do we overestimate content's impact?"
Distribution > Creation. Every time.
For fellow data scientists: This is why temporal validation matters. Train on 2024, test on 2025 revealed the Instagram algorithm shift. Random split would've hidden it.
Always test on future data for social media models.
Always test on future data for social media models.
November 3, 2025 at 3:00 AM
For fellow data scientists: This is why temporal validation matters. Train on 2024, test on 2025 revealed the Instagram algorithm shift. Random split would've hidden it.
Always test on future data for social media models.
Always test on future data for social media models.
The model's "failure" was its success. ROC-AUC 0.62 isn't bad at prediction - it's honest about reality.
Sometimes ML's greatest value isn't in what it predicts, but in what it proves isn't predictable.
Sometimes ML's greatest value isn't in what it predicts, but in what it proves isn't predictable.
November 3, 2025 at 3:00 AM
The model's "failure" was its success. ROC-AUC 0.62 isn't bad at prediction - it's honest about reality.
Sometimes ML's greatest value isn't in what it predicts, but in what it proves isn't predictable.
Sometimes ML's greatest value isn't in what it predicts, but in what it proves isn't predictable.
What this actually means:
Stop optimizing:
Perfect posting times
Ideal hashtag counts
Caption length formulas
Start optimizing:
Follower growth
Platform selection
Consistency over perfection
Stop optimizing:
Perfect posting times
Ideal hashtag counts
Caption length formulas
Start optimizing:
Follower growth
Platform selection
Consistency over perfection
November 3, 2025 at 3:00 AM
What this actually means:
Stop optimizing:
Perfect posting times
Ideal hashtag counts
Caption length formulas
Start optimizing:
Follower growth
Platform selection
Consistency over perfection
Stop optimizing:
Perfect posting times
Ideal hashtag counts
Caption length formulas
Start optimizing:
Follower growth
Platform selection
Consistency over perfection
This reminded me of Sutskever's detective analogy for LLMs - "predict the killer from the clues."
But in social media, the killer was decided before entering the room. The platform and follower count predetermined 70% of the outcome.
youtu.be/c2UGZmmgd0g
But in social media, the killer was decided before entering the room. The platform and follower count predetermined 70% of the outcome.
youtu.be/c2UGZmmgd0g
Predicting the Next Word - ilya sutskever - NVDIA AI Talk
Ilya Sutskever talks about Predicting the Next Word in Large Language Models
www.youtube.com
November 3, 2025 at 3:00 AM
This reminded me of Sutskever's detective analogy for LLMs - "predict the killer from the clues."
But in social media, the killer was decided before entering the room. The platform and follower count predetermined 70% of the outcome.
youtu.be/c2UGZmmgd0g
But in social media, the killer was decided before entering the room. The platform and follower count predetermined 70% of the outcome.
youtu.be/c2UGZmmgd0g
Platform-specific performance:
X/Twitter: Model caught 74% of high performers
Instagram: 0% (literally none to catch)
Threads: 30% recall
Same model. Same approach. Completely different results.
X/Twitter: Model caught 74% of high performers
Instagram: 0% (literally none to catch)
Threads: 30% recall
Same model. Same approach. Completely different results.
November 3, 2025 at 3:00 AM
Platform-specific performance:
X/Twitter: Model caught 74% of high performers
Instagram: 0% (literally none to catch)
Threads: 30% recall
Same model. Same approach. Completely different results.
X/Twitter: Model caught 74% of high performers
Instagram: 0% (literally none to catch)
Threads: 30% recall
Same model. Same approach. Completely different results.
The Instagram collapse:
2024: 3-5% average engagement
2025: 0.2% average engagement
Not a single Instagram post in 2025 qualified as "high performing" by my model's standards. The platform changed. The content didn't.
2024: 3-5% average engagement
2025: 0.2% average engagement
Not a single Instagram post in 2025 qualified as "high performing" by my model's standards. The platform changed. The content didn't.
November 3, 2025 at 3:00 AM
The Instagram collapse:
2024: 3-5% average engagement
2025: 0.2% average engagement
Not a single Instagram post in 2025 qualified as "high performing" by my model's standards. The platform changed. The content didn't.
2024: 3-5% average engagement
2025: 0.2% average engagement
Not a single Instagram post in 2025 qualified as "high performing" by my model's standards. The platform changed. The content didn't.
Feature importance was depressing:
has_numbers: 11%
video vs photo: 10%
hashtag_count: 6%
caption_length: 5%
posting_hour: 4%
Even the "best" content features barely matter.
has_numbers: 11%
video vs photo: 10%
hashtag_count: 6%
caption_length: 5%
posting_hour: 4%
Even the "best" content features barely matter.
November 3, 2025 at 3:00 AM
Feature importance was depressing:
has_numbers: 11%
video vs photo: 10%
hashtag_count: 6%
caption_length: 5%
posting_hour: 4%
Even the "best" content features barely matter.
has_numbers: 11%
video vs photo: 10%
hashtag_count: 6%
caption_length: 5%
posting_hour: 4%
Even the "best" content features barely matter.
The math is brutal:
~70% of engagement variance = which account/platform (distribution)
~30% = actual content quality
Your follower count matters more than your writing. By a lot.
~70% of engagement variance = which account/platform (distribution)
~30% = actual content quality
Your follower count matters more than your writing. By a lot.
November 3, 2025 at 3:00 AM
The math is brutal:
~70% of engagement variance = which account/platform (distribution)
~30% = actual content quality
Your follower count matters more than your writing. By a lot.
~70% of engagement variance = which account/platform (distribution)
~30% = actual content quality
Your follower count matters more than your writing. By a lot.
First red flag: When I included account/platform features, the model hit 0.77 ROC-AUC.
When I removed them to focus on pure content? Dropped to 0.62.
That 20% drop = the entire game.
When I removed them to focus on pure content? Dropped to 0.62.
That 20% drop = the entire game.
November 3, 2025 at 3:00 AM
First red flag: When I included account/platform features, the model hit 0.77 ROC-AUC.
When I removed them to focus on pure content? Dropped to 0.62.
That 20% drop = the entire game.
When I removed them to focus on pure content? Dropped to 0.62.
That 20% drop = the entire game.
Setup: I had posts from X, Instagram, Threads, LinkedIn, Facebook, YouTube. All my @Periospot content.
Built an XGBoost model with every feature I could think of:
Posting time
Hashtags
Caption length
Emojis
Content type
Goal: Predict high-performing posts
Built an XGBoost model with every feature I could think of:
Posting time
Hashtags
Caption length
Emojis
Content type
Goal: Predict high-performing posts
November 3, 2025 at 3:00 AM
Setup: I had posts from X, Instagram, Threads, LinkedIn, Facebook, YouTube. All my @Periospot content.
Built an XGBoost model with every feature I could think of:
Posting time
Hashtags
Caption length
Emojis
Content type
Goal: Predict high-performing posts
Built an XGBoost model with every feature I could think of:
Posting time
Hashtags
Caption length
Emojis
Content type
Goal: Predict high-performing posts
Tried to predict viral posts with machine learning on 5 years of my social media data.
The model failed spectacularly (ROC-AUC 0.62).
That failure taught me more about social media than any success could have.
Here's what 3,745 posts revealed
The model failed spectacularly (ROC-AUC 0.62).
That failure taught me more about social media than any success could have.
Here's what 3,745 posts revealed
November 3, 2025 at 3:00 AM
Tried to predict viral posts with machine learning on 5 years of my social media data.
The model failed spectacularly (ROC-AUC 0.62).
That failure taught me more about social media than any success could have.
Here's what 3,745 posts revealed
The model failed spectacularly (ROC-AUC 0.62).
That failure taught me more about social media than any success could have.
Here's what 3,745 posts revealed
I built an ML model to predict viral posts.
It “failed”… and then confessed.
Pure content → AUC 0.62
+ Distribution (platform/profile) → AUC 0.77
The 70/30 rule of virality:
~70% engagement = distribution
~30% = creative
Write-up + code: open.substack.com/pub/tuminha...
It “failed”… and then confessed.
Pure content → AUC 0.62
+ Distribution (platform/profile) → AUC 0.77
The 70/30 rule of virality:
~70% engagement = distribution
~30% = creative
Write-up + code: open.substack.com/pub/tuminha...
The Perfect Crime of Viral Content: What Ilya Sutskever’s Detective Tale Taught Me About Social Media
3,745 posts, temporal validation, and the rigged game of reach behind engagement.
tuminha.substack.com
November 2, 2025 at 2:00 PM
I built an ML model to predict viral posts.
It “failed”… and then confessed.
Pure content → AUC 0.62
+ Distribution (platform/profile) → AUC 0.77
The 70/30 rule of virality:
~70% engagement = distribution
~30% = creative
Write-up + code: open.substack.com/pub/tuminha...
It “failed”… and then confessed.
Pure content → AUC 0.62
+ Distribution (platform/profile) → AUC 0.77
The 70/30 rule of virality:
~70% engagement = distribution
~30% = creative
Write-up + code: open.substack.com/pub/tuminha...
I thought I was upgraded on GitHub, but it’s just Halloween season celebration colors.
November 1, 2025 at 8:55 AM
I thought I was upgraded on GitHub, but it’s just Halloween season celebration colors.
She's a 10, but she uses random_state=42
September 19, 2025 at 2:17 PM
She's a 10, but she uses random_state=42
I would sh** my pants if I were there.
September 5, 2025 at 9:30 AM
I would sh** my pants if I were there.
Doesn't matter how many times I see this; it hits me hard every time.
September 4, 2025 at 11:00 PM
Doesn't matter how many times I see this; it hits me hard every time.
I don't know how I could live without @PerplexityComet
Thanks, @RussellKirkDDS, for the codes!
Thanks, @RussellKirkDDS, for the codes!
September 4, 2025 at 3:04 PM
I don't know how I could live without @PerplexityComet
Thanks, @RussellKirkDDS, for the codes!
Thanks, @RussellKirkDDS, for the codes!
I was not aware of this tip to use Grok for proofreading on X 🤯
September 2, 2025 at 1:00 PM
I was not aware of this tip to use Grok for proofreading on X 🤯
