51% of companies have started using AI

Tech giants have cut over 53,000 jobs in 2025 itself

And 40% of professionals fear that AI will take away their job.

But here’s the real picture — companies aren't simply eliminating roles, they're hiring people who are AI-skilled, understand AI, can use AI & even build with AI.

Join the online 2-Day LIVE AI Mastermind by Outskill - a hands-on bootcamp designed to make you an AI-powered professional in just 16 hours.

Usually $895, but for the next 48 hours you can get in for completely FREE!

In just 16 hours & 5 sessions, you will:

✅ Learn the basics of LLMs and how they work.
✅ Master prompt engineering for precise AI outputs.
✅ Build custom GPT bots and AI agents that save you 20+ hours weekly.
✅ Create high-quality images and videos for content, marketing, and branding.
✅ Automate tasks and turn your AI skills into a profitable career or business.

🧠Live sessions- Saturday and Sunday
🕜10 AM EST to 7PM EST

All by global experts from companies like Amazon, Microsoft, SamurAI and more. And it’s ALL. FOR. FREE. 🤯 🚀

🎁 You will also unlock $5100+ in AI bonuses: 💬 Slack community access, 🧰 top AI tools, and ⚙️ ready-to-use workflows — all free when you attend!

Sponsored

Subscribe|Submit a tip|Advertise with Us

Welcome toDataPro 150: Your Weekly Brief on Data & AI 🚀

The pace of change in data and AIisn’tslowing down, and this week brings some of the most practical and forward-looking updates yet. From universal models and secure agent-led payments to tutorials you can run inColab, DataPro 150 is packed with the stories, tools, and insights that will shape your workflows.

Here are the highlights worth your time:

Buildanend-to-end voice AI agentwith Hugging Face pipelines onColab, combining Whisper, FLAN-T5, and Bark for real-time conversations.

ExploreMeta AI’sMapAnything, a transformer-based universal model for 3D reconstruction across 12 tasks, fully open-sourced.

Learn whyyourA/B test “winner”might be random noise, and how to design experiments withreal statisticalrigor.

See howGoogle’sData Science Agentnow integratesBigQueryML,DataFrames, and Spark to accelerate analytics with natural prompts.

Discover theAgent Payments Protocol (AP2), Google’s open standard for secure agent-led transactions backed by 60+ partners.

TryHugging FaceTrackio, a lightweightColab-native dashboard for experiment tracking and hyperparameter sweeps.

Plenty more awaits inside, from deep dives on retail sales shift analysis andFirestore’snew MCP tools, to hands-on coding with Zarr, advanced neural agents, and interpretable DNA CNNs. IBM’s Granite-Doclingalso makes a splash in document AI, while gradient boosted trees and unified intent recognition get the visual and structural treatment they deserve.

Together, these stories capture where AI is heading, smarter agents, more robust evaluation, and unified frameworks that bridge research and enterprise.

Let’sdive in. 🌊

Cheers,

Merlyn Shelley

Growth Lead, Packt

As a data professional, you already know how to find insights in complex information. But often, those insights stay in reports instead of powering real decisions.

That is where algorithmic trading comes in. It is the perfect add-on skill, taking what you already do with data and applying it to the financial markets.

On September 27, join Jason Strimpel, author of Python for Algorithmic Trading Cookbook, for a 2.5-hour live workshop where you will:

✅ Prototype and validate strategies with pandas
✅ Backtest the right way using VectorBT
✅ Deploy live systems with the Interactive Brokers API

💡 Plus, you will get: a free eBook, 90-day replay access, and a participation certificate.

⬛How to Build an Advanced End-to-End Voice AI Agent Using Hugging Face Pipelines?This tutorial explains how to build an advanced voice AI agent using Hugging Face pipelines on GoogleColab. It combines Whisper for speech recognition, FLAN-T5 for reasoning, and Bark for speech synthesis, avoiding APIs or heavy dependencies. The guide covers transcription, response generation, speech synthesis, conversation management, and aGradioUI for real-time interactive voice conversations.

⬛Meta AI Researchers Release MapAnything:MapAnythingis a transformer-based universal model for 3D reconstruction that supports over 12 tasks such as monocular depth, multi-view stereo, and structure from motion in a single feed-forward system. Built on DINOv2 features with a factored scene representation, it processes up to 2,000 images with optional priors, achievesstate-of-the-artresults, and isopen-sourcedunder Apache 2.0 with complete training resources. This blog explores its architecture, training strategy, benchmarks, and key contributions.

⬛Why Your A/B Test Winner Might Just Be Random Noise?An 8% boost in sprint speed sounds like a breakthrough, but it might just be chance. This post explores how randomness can mislead us in A/B tests, illustrated through a football team’s warm-up experiment. By unpacking the pitfalls of small samples and uncontrolled factors, it shows how proper design, replication, and statistical rigor separate real signal from noise.

⬛Data Science Agent now supports BigQuery ML, DataFrames, and Spark:Google is bringing an AI-firstColabEnterprise notebook experience to Vertex AI, designed to simplify and accelerate data science workflows. This blog explores how theData Science Agentnow supportsBigQueryML,BigQueryDataFrames, and Spark generation from prompts, adds context-aware data retrieval and @ mentions, and enables seamless automation of data exploration, transformation, and modeling at scale.

⬛Announcing Agent Payments Protocol (AP2):Google introduces the Agent Payments Protocol (AP2), an open standard for secure agent-led transactions that extends A2A and MCP. AP2 uses cryptographically signed Mandates and verifiable credentials to prove intent, authorize carts, and create an auditable trail. It supports cards, bank transfers, and stablecoins, is backed by 60+ partners, enables new commerce flows, and ships with public specs and reference implementations.

⬛A Comprehensive Coding Guide to Building Interactive Experiment Dashboards with Hugging Face Trackio:This tutorial walks through Hugging FaceTrackiofor clean, local experiment tracking in a singleColabnotebook. You installTrackio, build a synthetic dataset, run multiple SGD training configs, and log metrics and confusion-matrix tables. A small hyperparameter sweep summarizes best settings, results import from CSV, and the lightweight dashboard updates in real time, giving intuitive visibility into runs and performance.

⬛Analysis of Sales Shift in Retail with Causal Impact:Estimating how sales shift when a product disappears from shelves is a complex but crucial task for retailers. This article explores Carrefour’s use of Google’sCausal Impactmethod, whichleveragesBayesian structural time-series models to build synthetic controls. It explains the use case, strategies for handling anomalies, covariate selection, model design, and validation to produce reliable estimates of lost and transferred sales.

⬛Firestore support and custom tools in MCP Toolbox:MCP Toolbox for Databases is an open-source server that connects AI agents to enterprise data, with support forBigQuery,AlloyDB, Cloud SQL, and Spanner. This article introduces newFirestoretools that bring AI-assisted development to the NoSQL world. From querying documents and cleaning data tovalidatingsecurity rules, developers can now manageFirestoredirectly through natural language in environments like Gemini CLI.

⬛A Coding Guide to Implement Zarr for Large-Scale Data:This tutorial exploresZarr, a library for efficient storage and manipulation of large multidimensional arrays. Starting with array creation, chunking, and on-disk edits, it moves into advanced operations like compression benchmarks, hierarchical dataset structures, time-series simulations, and volumetric indexing. You also learn chunk-aware processing and data visualization, gaining hands-on experience with Zarr’s performance, scalability, and flexibility for real-world scientific workflows.

⬛How to Build a Robust Advanced Neural AI Agent with Stable Training, Adaptive Learning, and Intelligent Decision-Making?This tutorialdemonstrateshow to design and implement anAdvanced Neural Agentthat blends classical neural network methods with modern stability techniques. It covers Xavier initialization, stable activations, gradient clipping, momentum updates, and weight decay. The training loop integrates mini-batching, adaptive learning rates, early stopping, and instability resets. Extended with experience replay and exploratory decisions, the agent adapts to regression, classification-to-regression, and RL-style tasks.

⬛ROC AUC Explained: A Beginner’s Guide to Evaluating Classification Models.Evaluating binary classification on imbalanced datasets requires morethan accuracyalone. In the IBM HR Analytics case, logistic regression reached 86%accuracy, yetrecall for employees who left was just 34%. This gap highlights why ROC AUC is essential. By analyzing true positive and false positive rates across all thresholds, it provides a balanced, threshold-independent measure of model quality.

⬛Automate app deployment and security analysis with new Gemini CLI extensions:Close the gap between terminal and cloud with Gemini CLI’s new extensions. The security extension adds/security:analyzefor local vulnerability scans with actionable fixes and upcoming GitHub PR reviews. The Cloud Run extension adds/deployto build and ship apps to a public URL in minutes via an MCP-backed pipeline. Install the extensions, authenticate withgcloud, and deploy or scan from one place. This blog is about simplifying secure development and deployment workflows directly from Gemini CLI.

⬛IBM AI Releases Granite-Docling-258M:IBM has introduced Granite-Docling-258M, an open-source vision-language model built for end-to-end document conversion. It improves overSmolDoclingwith a Granite 165M backbone, SigLIP2 vision encoder, and stability fixes, achieving higher accuracy in layout, OCR, tables, code, and equations. EmittingDocTagsfor structured output, it supports multilingual text, integrates withDoclingpipelines, and runs efficiently across runtimes. This blog is about advancing enterprise-ready, structure-preserving document AI.

⬛Building an Advanced Convolutional Neural Network with Attention for DNA Sequence Classification and Interpretability:An advanced convolutional neural network can be built to classify DNA sequences by combining one-hot encoding, multi-scale convolutional layers, and attention for interpretability. This tutorial walks through generating synthetic data, training with callbacks, and visualizing results across promoter prediction, splice site detection, and regulatory element tasks. The workflowdemonstrateshow deep learning can capture biological motifs while offering transparency. This blog is about applying CNNs with attention to DNA sequence classification in a reproducible, interpretable way.

⬛Building a Unified Intent Recognition Engine:Intent recognition often sits in silos across enterprise teams, each building bespoke pipelines for chatbots, triage tools, or assistants. A unified approach simplifies this by standardizing reusable steps, preprocessing, embeddings, vector search, and scoring, while allowing project-specific customization. The Unified Intent Recognition Engine (UIRE) accelerates deployment, reduces redundancy, and supports advanced features like multi-intent detection and out-of-scope handling. This blog is about creating a modular, scalable framework for enterprise-wide intent recognition.

⬛A Visual Guide to Tuning Gradient Boosted Trees:Gradient boosted trees extend decision trees and random forests by building trees sequentially, each correcting the errors of thepreviousones. Using scikit-learn for visualization, we can see predictions refine over iterations, errors shrink, and performance shift with hyperparameters like learning rate, depth, and estimators. This exploration highlights their strengths, trade-offs, and practical behavior in real-world applications.

See you next time!