Tobias Zwingmann

RAPYD.AI

Independent AI consulting to create profitable AI solutions to innovate and grow your business. Fast.

Managing Partner

O'Reilly Media

Author and Instructor

LinkedIn

LinkedIn Learning Instructor

D2A2

Chief of Research - Applied AI

Springboard

- Helping Springboard as a subject matter expert in AI, especially for data analytics
- Mentoring students who want to switch careers to data science / data analytics 
- Top Mentor Award in 2021 (Springboard Inspire)

AI SME & Data Science Mentor

Insurgence

Partner

Toptal

Independent Consultant

Data Science Dojo

ML/AI Instructor

Deutsche Messe AG

Systematic retrieval, analysis and utilization of (unstructured) data from diverse sources to build new products and processes.

- People movement analytics 
- Computer vision with focus on text retrieval

Senior Data Scientist

Building & maintaining predictive models. Generating and communicating insights about customer trends and behavior to product owners and senior management.

- Churn Prediction
- Next Best Offer Recommendations 
- Customer Segmentation
- Visual Analytics

Data Scientist

Leading various international trade show projects in e.g. China, India, Indonesia, Mexico, Myanmar, South Africa, South Korea, Turkey.

Project Director

Deutsche Messe Dubai Branch

Management Trainee

Author of AI-Powered BI (O'REILLY) | AI Coach & Mentor | I help organizations leverage AI and use data more intelligently to improve business performance. Sharing AI use cases & learnings.

I didn’t start my AI/ML journey doing “Profitable AI”.

I started it as a Data Scientist who built models that looked impressive on paper:

– Customer churn prediction with 95% accuracy ✅
– Next-best-offer systems with 15% uplift ✅
– Data-driven B2B products generating 6-figure revenues from day one ✅

But the hard lesson I’ve learned after 10+ years at the intersection of business and technology is that none of this matters if you can’t show results on the bottom line.

I’ve spent the last 5+ years in applied AI consulting where my work has taken me inside global enterprises in automotive and insurance, mid-sized companies with 500–1,000 employees, solopreneurs aiming to double their output, and small business owners uncovering $10k+ monthly savings in unexpected places. 

The contexts differ, but the patterns of what makes AI succeed (and fail) remain remarkably consistent.

Success is not about:
- Chasing the next shiny AI tool
- Hiring data scientists from Silicon Valley
- Hoping IT will figure it out

It’s about applying the principles of Profitable AI to deliver outcomes that matter: doing things faster, at lower cost, with higher quality — or scaling what you do to entirely new levels.

Along the way, I’ve shared many of these lessons in my books AI-Powered Business Intelligence (O’Reilly, 2022) and Augmented Analytics (O’Reilly, 2024), as well as in my weekly newsletter that is now read by 10,000+ professionals at companies like Google, Amazon, Gucci, H&M, Santander, and Mercedes-Benz.

So If you’re tired of the shiny AI demos and want to see it deliver real results instead — I think you found a good place.

We help you build amazing AI products faster.

Gemini 3 is making a lot of noise. And it's a really great model. "Is it better than GPT-5?!" My take: Use whatever works for you. Both are LLMs, both have their quirks. But if you want to know where each shines, here's what I found:

Gemini 3 = Messy Inputs → Clean Outputs
Throw it chaos: screenshots, transcripts, logs, mixed documents. It's great at compressing that mess into structure. Best when you're very specific about the output format you want. Seems like if you need to organize messy information, call Google.

Examples:
- Summarizing multiple meeting recordings into structured notes
- Extracting data from a messy pile of PDFs, screenshots, and spreadsheets

GPT-5.1 = Clean Inputs → Complex Thinking
Give it organized information and let it reason. It's great at planning, narrative, and multi-step decisions. Best when the task itself is complex, not the input.

Examples:
- Writing code (though Claude still beats it here)
- Building a strategy doc from a clean brief

Scenarios to combine both:

1/ Competitive analysis
Gemini 3 gathers data from competitor sites, pricing, and reviews; GPT 5.1 delivers the strategy

2/ Client onboarding
Gemini 3 cleans and organizes intake docs; GPT 5.1 produces the project plan

3/ Sales pipeline review
Gemini 3 compiles CRM data, calls, and emails; GPT 5.1 spots trends and recommends actions.

Don't try to find "the" best. 
Find whatever works best for you.

Fun fact: The infographic was done by me with Gemini-3 image gen ("Nano Banana"), first shot, Prompt in comments.

AI Advisor | O’Reilly Author | Instructor @ LinkedIn Learning | I find & realize profitable AI opportunities – here to share wins, fails, examples & frameworks.

I see a lot of confusion around AI workflows, AI agents, workflow automation, and agentic workflows. Most people can't even tell the difference. Which isn't a problem until you want to build an AI solution that is neither over-engineered nor insufficient.

So I put together a quick guide that breaks down each type and when to use what, along with a few examples.

Read the full blog post here: https://lnkd.in/erYqeVPi

AI Workflows vs. AI Agents

ChatGPT launched today 3 years ago. ONLY three years I should say. Feels like a decade compressed. To catch you up on what’s happened, I generated this little timeline with the help of ChatGPT Deep Research, had Gemini 3 double-check it, and Claude 4.5 generated the PDF.

Tasks like this would have taken a whole day in the "old" world.
Now done in less than 60 minutes with AI.

We're still far away from AGI.
But I don't care.
We're living in the most exciting timeline.
And I can’t wait to see what this timeline looks like 3 years from now.

3 Years Since ChatGPT

A while ago, I shared the AI Profit Stack – a simple framework to assess whether your AI projects are actually creating profit. People loved it, clients used it. Then someone asked me:

“How do I apply this to my Copilot licenses?”

Short answer:
You don’t.

Long answer:
Trying to measure all AI with the same ROI formula is exactly how you end up with fictional numbers, cynical finance teams, and AI steering committees that debate themselves into extinction.

Today I've shared the fix I'm using now in every client engagement – and the one I want you to have too.

Find it here here:

Open article: The Two Tracks of AI ROI by blog.tobiaszwingmann.com

How to separate 'AI that helps' from 'AI that pays'

𝗚𝗘𝗢𝗙𝗜𝗧 𝗚𝗥𝗢𝗨𝗣

GEOFIT designs geographical databases to create industrial and infrastructure projects.

At the Innovation Lab (Geofit R&D unit) we work on multiple types of geospatial data mostly produced by the other units of the group such as HD drone pictures and point clouds from mobile and airborne LiDAR as well as open source satellite images.
We experiment with the most recent deep learning architectures for a variety of computer vision tasks.

ML Research Engineer

BreadcrumbsIO (YC S21)

Breadcrumbs is a revenue acceleration platform based on a co-dynamic lead scoring and routing engine living at the intersection of marketing and sales.

Developed and deployed an explainable model for predicting lead to customer conversion: the model is agnostically applied to imbalanced, heterogeneous and sparse data and it can capture relevant features allowing for an automated and data-driven lead scoring model pre-compilation

Kosmopolead

Software company specialized in CRM solution.

Developed a news content recommendation system using pre-trained word embedding, clustering and dimensionality reduction techniques

Full stack web development: developed mainly Ruby on Rails apps,  Gems and  RESTful APIs

WEB & software developer-researcher

Multiple certificates on machine learning and web development

Junior WEB & software developer-researcher (Parcours TPME - Pôle Emploi)

Nicolaus Copernicus Astronomical Center of the Polish Academy of Sciences

- Implemented Fourier analysis in my code MoCA to study the time variability of Comptonization in accreting SMBHs.
- Given talks and seminars and worked with international teams.

Scientific Collaboration

Astronomical Institute of the Czech academy of Sciences

- I coupled my radiative transfer Fortran code, MoCA, with a ray-tracing routine written in C in order to have a full GR description of the black hole accretion process.
- I was the main contributor to one of the working packages of the FP7 European research grant.
- Talks and seminars in international conferences.

Postdoctoral Research Fellow

Observatoire astronomique de Strasbourg

- Translated my code MoCA in Fortran for parallelism on clusters
- Implemented accurate Compton scattering in the STOKES code

Private teacher in mathematics and physics for university and high-school students

Private Teacher

GAMBERO ROSSO S.P.A.

Cook chef’s assistant for private events such as wine and food tasting and gala dinners.

Chef's assistant

OP&R srl

Waiter service for wine bars, restaurants, wine tasting and private dinner events.

Waiter

COPIT onlus

Factotum

Data scientist with clear analytical skills, I take advantage of my proven experience in scientific research, statistics and data analysis to convert complex problems into creative and original deployable solutions.

I think that AI is the key technology of this century and it's already shaping our lives and society. In the next couple of decades it's going to be either an invaluable tool in view of the inevitable change of economic ideology we are going to accomplish all together for the survival of our species (willingly or not), or it's going to be integrated into products and services with more and more of our biases favouring political polarization and division for the profit of a few.
I believe that with the correct objective function and if used ethically and applied responsibly, AI can help humanity converge towards a common path of prosperity, equity and sustainability.

As a former theoretical astrophysicist I developed from scratch "MoCA", a Monte Carlo code to study black holes physics. My numerical simulations involved advanced statistical analysis of hundreds of millions of simulated photons and at that time MoCA was a state-of-the-art code used by different teams in international collaborations and projects (including assessment study reports for ESA/NASA space missions).


STACK:

Languages: 
   🐍 Python
   🔍 SQL
   🧑‍💻 Fortran
    🕸 Web (HTML, CSS, JS)
   💎 Ruby
   📜 LaTex
 
ML frameworks / libraries:
  🩸 PyTorch
  🎛 Keras TensorFlow
  🤖 Scikit-learn
 
Engineering:
  🌍 GIS (Rasterio, Geopandas, Shapely, QGIS, CloudCompare, Folium, STAC)
  ⚙ MLOps (MLFlow, wandb, Valohai, Sagemaker)
  🐈‍⬛ Git, GitLab
  🐳 Docker
  ☁ AWS, GCP

Web frameworks:
  ⚡ FastAPI
  🧪 Flask
  🛤 Rails

taplio

The Battle

Tobias Zwingmann

Francesco Tamborra 🖖

Followers

Posts

Engagements

Likes

Performance ScoreTobias Zwingmann's LinkedIn Performance

Ready to grow your LinkedIn presence?

taplio

The Battle

Tobias Zwingmann

Francesco Tamborra 🖖

Followers

Posts

Engagements

Likes

Performance Score.css-goub6s{font-size:var(--chakra-fontSizes-md);font-weight:500;color:var(--chakra-colors-gray-600);display:block;margin-top:var(--chakra-space-1);}Tobias Zwingmann's LinkedIn Performance

Ready to grow your LinkedIn presence?

Performance ScoreTobias Zwingmann's LinkedIn Performance