Prototyping Custom AI enterprise Solutions

In today’s boardrooms, decisions are increasingly influenced by digital transformation and the ability to translate data into strategy. Yet many enterprises still struggle to bridge the gap between business needs and practical AI implementations.

In this demo-driven session, I will showcase how Python can be used to rapidly prototype custom AI workflows tailored to enterprise contexts. Using a social media sentiment analysis prototype as the case study, we will walk through framing the business problem, building the proof-of-concept with Python libraries, and delivering outputs that matter to leadership—such as daily sentiment breakdowns, risk theme detection, and exportable executive summaries.

Attendees will learn a practical approach to moving from idea → prototype → business impact using open-source tools, with examples they can adapt for their own organizations.

Format: Talk + Live Demo

Audience Takeaways:

1. How to translate business pain points into Python AI prototypes

2. A practical workflow for experimenting with sentiment analysis in an enterprise setting

3. Lessons on communicating AI insights to non-technical leadership

Behind the scenes, this demo runs on Python as the main programming language. It connects to a "database" or uses a file as the data source. It uses Google’s Gemini and PaLM AI models to analyse and generate text, and links with Gmail to automatically send out results and reports.

Preview of the slides: https://docs.google.com/presentation/d/1TufvOnWPHHKoB9tUXSNcj1CV_nnB680krLcBNwUNoF4/edit?usp=sharing

Créer des applications impactantes avec l'IA et le cloud – Retour d'expérience

Cette présentation s'inspire de mon parcours personnel : une transition du métier de Data Scientist vers celui d'Ingénieur en IA, en utilisant l'Intelligence Artificielle comme compagnon de développement au quotidien. Je vous montrerai, à travers deux exemples concrets, comment le cloud computing permet de surmonter des contraintes matérielles comme l'absence d'ordinateur portable ou une puissance de calcul limitée.

En m'appuyant sur deux outils Project IDx et Google AI Studio , je vous expliquerai comment :

1. Créer des scripts pour automatiser les tâches répétitives,
2. Développer des prototypes opérationnels afin de valider rapidement une idée.

Vous découvrirez comment ces plateformes, combinées à Gemini, transforment des concepts abstraits en applications concrètes et impactant

Thriving Like a True Dev Ninja!

A talk about where most developers have been, a space where self-doubt and error challenge your mental well-being. join me in an electrifying discussion where we'll explore the profound impact of code on our sanity. Picture this: you're drowning in a sea of meaningless commits, or worse yet, you unleash an unholy production error that defies all logic. I will equip you with invaluable insights and ingenious ideas to conquer the challenges that plague us when writing, debugging, and committing code. Trust me, these tips and tricks have kept me from the brink of madness, managing the treacherous terrains of mental fatigue, information overload, and the infamous burnout.

Building Production-Ready GenAI Apps in Java with Vertex AI

We have all seen the flashy GenAI demos but what happens when you actually need to run one in production?

In this session, I will share how I built a Java service on Vertex AI that takes real business data (Google Analytics + revenue goals) and turns it into useful customer personas for the marketing department.

I’ll walk you through the code and configs step by step:

1. how I structure prompts to keep the model grounded in context,

2. what Vertex AI settings really matter (project, region, model IDs),

3. the service layer code that ties everything together, and

4. how I packaged and deployed it on Cloud Run with secrets, scaling, and monitoring.

The slides: https://docs.google.com/presentation/d/1REjA5N90YXSou4YeovhxeDrco7LL0Godeky6Ua0M3CE/edit?usp=sharing

Dev can❤️ Marketing: Building a Brand-Safe Copy Assistant

Developers dread “quick copy tweaks.” Marketers dread “it’s on the backlog.” This talk proposes a truce: a tiny, production-lean Multilingual Ad Copy Assistant that turns briefs into on-brand variants, localizes for Shona, Ndebele, and French, enforces a brand glossary, and runs policy checks all as a Cloud Run API your team can actually live with.

Why devs won’t hate it: one endpoint, predictable JSON, idempotent requests, glossary as code, and a deploy in minutes.

Why marketing will love it: faster variants, consistent voice across markets, and fewer “no, we can’t say that” moments.

A Developer’s Field Guide to Social Good

A Real Talk on Building for Impact, derived from my experience as a Women Techmakers Ambassador and developer working on a project for social good with the Internet Society of ZImbabwe.

This session explores how low-code tools can be used to support large-scale, real-world initiatives in under-resourced settings. Drawing from her experience as a developer supporting a national gender digital inclusion study across Zimbabwe, it demonstrates how platforms like Google AppSheet and Looker Studio enabled rapid deployment of data collection and visualization tools without requiring extensive infrastructure.

The talk walks through the creation of an offline-capable mobile application used by over 20 field enumerators and the development of a real-time dashboard for visualizing regional progress, digital literacy scores, and access disparities.

Attendees will learn:

1. How to build mobile-first, offline-ready data collection tools using AppSheet

2. How to connect Google Sheets and Forms for live data syncing

3. How to use Looker Studio to design dashboards for stakeholder engagement

4. Techniques for managing user permissions and version control in low-bandwidth contexts

5. Ways to translate raw field data into insights that support policy and funding conversations

This session is aimed at developers and technologists interested in using familiar tools to drive social impact. I

Explore the article and visuals from this project on the link: https://www.linkedin.com/pulse/how-developers-shape-impact-ruvimbo-delia-hakata-x9ubf

Prototyping Advertisement Concepts with Google TTS

Great creative work begins with play and culminates in outcomes. In this session, I share a lightweight workflow for turning a marketing idea into a first-draft campaign asset in minutes.

Using Google Colab, Google Cloud Text-to-Speech, and a simple audio toolchain, I will demonstrate two prototypes: a single-voice and two-voice “duet” advertisement built from prompts. We will examine where artificial intelligence accelerates work (drafting scripts, generating variations, and producing voice lines) and where humans must finish the job (message strategy, tone, and cultural nuance).

I will also outline the next-step roadmap: connecting Google Analytics to guide scripts with real audience interests, so drafts begin closer to what will perform.

Who can be interested: Marketers, social media managers, and creatives who want faster first drafts without losing authenticity.

You will learn:

-- A prompt-to-prototype pipeline (Colab → Text-to-Speech → duet mix).

-- How to reduce iterations while preserving brand voice.

-- A practical path to analytics-informed creative.

I have detailed part of the demo in an article which can be accessed on this link: https://www.linkedin.com/pulse/enhancing-presentations-how-ai-tools-generate-campaign-hakata-lrnwf

Offline AI for Africa’s Enterprise Stack

Not everyone needs to send data to the public cloud to benefit from large language models. In this session, I demonstrate how to embed private LLM inference directly into existing Java-based systems using Ollama and Gemma3.

We will walk through an offline demo that uses a JSON “sales” dataset, generates prompts in Java, and returns results locally no internet connection and no data leaving your environment. I will explain why Java remains a practical choice for mission-critical backends across Africa (and beyond) and how Gemma3’s architecture (including grouped-query attention and RMSNorm) balances speed with accuracy for real-time queries.

This approach is especially relevant for regulated industries that must keep sensitive data private such as healthcare, banking, insurance, and retail loyalty programs handling personally identifiable information. Many teams in these sectors face strict data residency rules, confidentiality requirements, and audit obligations. By running inference locally, organisations can preserve privacy, reduce exposure risk, and maintain control over logs and model inputs while still gaining AI-level capabilities.

Who it is for:

Enterprise developers, solution architects, and product leaders in regulated or privacy-sensitive environments who need AI capabilities on existing stacks and must keep data private.

What attendees will learn

-- How to run private LLM inference locally with Java, Ollama, and Gemma3.

-- A repeatable offline workflow: JSON data → Java prompts → local inference → outputs that can feed terminal, reports, or a user interface.

-- Where private inference fits today in regulated industries, and how to extend it with Retrieval-Augmented Generation (RAG) tomorrow.

-- Practical considerations for latency, reliability, auditability, AI ethics and data protection in real-world environments.

I have delivered this talk for GDG Harare's Build with AI Workshops and written an article that outlines what the workshop will include. You can access the article here: https://www.linkedin.com/pulse/private-llm-inference-enterprises-ruvimbo-delia-hakata-tvrjf