Build With AI - In Person Session Slides.pdf

Build with
AI during
#BuildwithAI
Vibe coding with Firebase Studio

What is the vibes???
Editable Location

Editable University Name
Prepare and understand
Get on to Firebase Studio
Prompt our first app & enter API key
Get wowed , disappointed and remember that AI is a tool
Redo until it works (might be forever though)
Our todo list

Chapter One
Getting everything ready and set
Setting the vibes
right.

The general who wins the battle
makes many calculations in his
temple before the battle is fought.
The general who loses makes but few
calculations beforehand..
- Sun Tzu

Our Step 1
Understand
your
requirements
A very good
prompt
A fire playlist
Internet of
course
Patience
Step 1
Basic functionality Basic UI Some errors

Chapter Two
Let AI generate the vibes
Generating our
first web app

Every first draft is
perfect, because all a
first draft has to do is
exist.
- Jane Smiley

Chapter Three
Let us control the vibes
Making edits to
our web app

Some people call this artificial
intelligence, but the reality is this
technology will enhance us. So
instead of artificial intelligence, I think
we'll augment our intelligence
Ginni Rometty, Former
CEO of IBM

Inside the AI Lens:
How AI Sees, Understands,
and Edits the World
Harare
Michael Nyamande

Cameras can see, but they can’t understand.
AI changes that. It’s like giving a camera a brain
— one that doesn’t just look at the world, but
interprets it, reshapes it, and even imagines new
versions of it.

● Software Developer
About Me
● Community Builder
● Technical Writer
● Author
● AI Enthusiast

● Software Developer
About Me
● Community Builder
● Technical Writer
● Author
● AI Enthusiast
Zim Tech Community

In this session, we’ll
walk through how AI
models like Imagen 3
and Gemini 2.5 enable
us to perform Virtual
Try-On
What we’ll build

Think of it as teaching AI to spot things in an image —
“this is a person”, “that’s a chair”, “this is a shoe”. It's
like giving eyes to a computer.
🔍 Object Detection

While object detection finds where an object is,
segmentation goes deeper — it outlines the exact shape of
that object, pixel by pixel. It's like cutting out the object with
precision scissors.
🧩 Image Segmentation

Once we know where the object is, we create a mask —
like a stencil — so the model knows which parts of the
image to protect, edit, or replace.
🎭 Masking

Ever used the healing brush in Photoshop? Inpainting is
the AI version — it fills in or redraws parts of an image,
using surrounding context. Want to remove a shirt and
replace it with a hoodie? That’s inpainting.
🎨 Inpainting

💡 Beyond Try-On: Real-World Use Cases
The same building blocks behind this — segmentation, object
detection, masking, and inpainting — power a wide range of
image editing applications across industries.
What use cases can you think of ?

Source: xxxxx
Visualizing furniture
placement, room layout
changes.
Segmenting organs in
medical scans for diagnosis
and surgery planning.
🎨 Design & Architecture 🏥 Healthcare
Real-time segmentation to
detect pedestrians, roads,
and obstacles.
🚗 Self-driving Cars
Background editing, scene
enhancement, and CGI
support.
Restoring old, damaged
photos using AI inpainting.
🎬 Film & Gaming 📰 Journalism & History
Segmenting crops vs weeds,
land mapping via drones
🌱 Agriculture

Embedding & LLM
Inference in Enterprise
Systems
Build With AI Workshop - 10/05/2025
Harare

Your Big
Idea Needs
A Business
Use Case.
Hi, I am Ruvimbo Delia Hakata
an AI Engineer.

Which languages power
business core systems?
Harare

9.31%
Java maintains strong
global presence over C#
which is at 4.22%.
Based on the TIOBE Index May, 2025
Learnings - Languages
used Enterprise core
Systems
Java's strong presence in Africa reflects
the continent's mobile-first approach to
technology adoption, where Android
dominates the smartphone market. Java's
role in developing mobile applications
makes it particularly valuable in the
African context.

What is LLM
Inference ?
Chapter One
How is it used in enterprise.

At its core, LLM Inference is the process where a trained Large
Language Model (LLM) actually uses its knowledge to generate a
response or make a prediction based on new input it receives .Think
of an LLM as a very advanced AI system that has been "trained" on a
massive amount of text and code. This training allows it to understand
patterns, grammar, context, facts, and even some reasoning abilities
related to human language.
Inference is what happens after the model is trained. It's the "live"
phase where the model applies its learned knowledge. It's like the
trained student now taking an exam or answering questions in the real
world.
LLM Inference Is

● When you give an LLM a prompt (like asking a question, giving it
an instruction, or starting a sentence), the inference process
begins.
● The LLM takes your input. {Say Gemini 2.0}
● It processes this input through its complex network of "neurons"
and "parameters"
● It then generates an output, typically word by word or "token by
token" . It predicts the most probable next token, then the next,
and so on, to form a coherent and contextually relevant
response.
● The goal is to produce text that is helpful, informative, creative,
or whatever the specific task requires, based on the input
prompt.
The Inference Process

Chapter Two
What are
Embeddings ?

Embeddings are
the LLM's way of translating words and
phrases into a numerical language it
understands – a language where
distance means similarity in meaning. The magic of closeness
allows them to perform all sorts of
intelligent tasks by navigating this rich
'meaning map'.
LLM Embedding

Let’s Get Into The
Demo !
Harare

An Attendee Asked
As a programmer what is the system
that i can generate for supermarket like
Food lovers to help it using AI?
I asked them
Which part of their business are we
helping them with?
They said to create promotions
Back Story - From The Crowd

The Problem
Making fresh promo stuff for
supermarkets like Food Lovers can be a
real headache and super
time-consuming. As a programmer, I'm
thinking there has to be a smarter,
quicker way to get those deals
advertised. It's just tough to keep up
and make ads that really grab attention
with the usual grind.
Solution
Imagine an AI sidekick that helps Food
Lovers (and us programmers!) whip up
cool promotional material easily. AI could
dream up catchy ad text, suggest
designs, or even pick out hot products to
feature. This means awesome, targeted
promos without all the usual fuss, making
marketing way smoother.

code_slides.txt
// A couple more notes on what we are doing:
// We simulate the data in the database with the hard coded Json
// We call LLM API
// We are using Ollama for offline access {because businesses like
to keep their data private}
// We demo with Ollama on my localhost and the cloud
Code can be
found on GDG
Harare Repo

code_slides.txt
String promptText = basePrompt + dataArray.toString(2);
// Connect to local LLM inference endpoint
URL url = new URL("http://localhost:11434/api/generate");
HttpURLConnection conn = (HttpURLConnection)
url.openConnection();
conn.setRequestMethod("POST");
conn.setRequestProperty("Content-Type", "application/json;
utf-8");
conn.setDoOutput(true);
Use this
template for
code snippets
longer than
6 lines
Build prompt by
combining your
base template with
the JSON-encoded
sales data

code_slides.txt
String promptText = basePrompt + dataArray.toString(2);
// Connect to local LLM inference endpoint
URL url = new URL("http://localhost:11434/api/generate");
HttpURLConnection conn = (HttpURLConnection)
url.openConnection();
conn.setRequestMethod("POST");
conn.setRequestProperty("Content-Type", "application/json;
utf-8");
conn.setDoOutput(true);
POST your prompt
to the local LLM
endpoint

code_slides.txt
Read the JSON
response,
extract and
display the
response ﬁeld

Key Observations:
1. Item Pricing and Currency:
○ Sadza+Chicken: $2 USD (highest price).
○ Sadza+Beef Bones: $1 USD (lowest price
in USD).
○ Coleslaw: 1 ZWG (Zimbabwean Dollar, but
no exchange rate provided).
2. Potential Gaps:
○ Crucially, there's no data on sales volume
(e.g., units sold, frequency), which is
essential to understand item popularity
and demand.
○ The Coleslaw is priced in ZWG, potentially
indicating a different regional market or a
data entry issue (e.g., "ZWG" might be a
typo for "ZAR" or "USD").
Recommendation:
Promote "Sadza+Chicken" (priced at $2 USD).
Code_slides.txt
// The Input Data and the output

Build With AI - In Person Session Slides.pdf

More Related Content

Similar to Build With AI - In Person Session Slides.pdf (20)

Recently uploaded (20)

Build With AI - In Person Session Slides.pdf