Velamuvi AIAcademy

Project library

Build real AI projects

You learn AI by building. Each brief tells you what to make, the skills it uses, the courses that prepare you and the milestones to reach, with ideas for going further.

Beginner projects

First programs. All you need is basic Python.

Programming Foundations

Python Calculator

A command-line calculator that reads two numbers and an operator, then prints the result.

  • Variables
  • Input
  • Conditions
  • Functions
Milestones and next steps
  1. 1Read two numbers from the user
  2. 2Add the four basic operations
  3. 3Handle division by zero
  4. 4Wrap each operation in a function
  5. 5Loop until the user quits

Go further

  • + Add powers and square roots
  • + Keep a history of calculations

Prepare with

Programming Foundations

Number Guessing Game

The computer picks a secret number and gives higher or lower hints until you find it.

  • Loops
  • Conditions
  • Random numbers
Milestones and next steps
  1. 1Generate a random number
  2. 2Ask for a guess in a loop
  3. 3Give higher or lower hints
  4. 4Count attempts
  5. 5Offer to play again

Go further

  • + Add difficulty levels
  • + Let the computer guess your number with binary search

Prepare with

Data Science & Engineering

Basic Data Analysis

Load a real CSV dataset, clean it and answer five questions with Pandas and charts.

  • Pandas
  • Cleaning
  • Charts
Milestones and next steps
  1. 1Load and inspect the data
  2. 2Fix missing values and types
  3. 3Answer questions with groupby
  4. 4Chart each answer
  5. 5Write a short findings summary

Go further

  • + Build a reusable analysis notebook
  • + Add an interactive chart

Prepare with

Generative AI

AI Chatbot

A chat app that remembers the conversation and streams replies from an LLM API.

  • AI APIs
  • Prompting
  • Conversation history
Milestones and next steps
  1. 1Make a first API call
  2. 2Keep message history
  3. 3Write a system prompt
  4. 4Stream the response
  5. 5Handle errors and limits

Go further

  • + Add a web interface
  • + Summarise long history to save tokens

Prepare with

Intermediate projects

Real datasets and real models. Great portfolio pieces.

Machine Learning

House Price Prediction

Predict home prices from features such as size and location using regression.

  • Regression
  • Feature engineering
  • Evaluation
Milestones and next steps
  1. 1Explore the dataset
  2. 2Prepare features
  3. 3Train a linear baseline
  4. 4Try a tree-based model
  5. 5Compare with MAE and R²

Go further

  • + Tune with cross-validation
  • + Serve predictions through an API

Prepare with

Machine Learning

Spam Detection

Classify messages as spam or not spam using text features and Naive Bayes.

  • Text features
  • Classification
  • Precision and recall
Milestones and next steps
  1. 1Load labelled messages
  2. 2Turn text into TF-IDF vectors
  3. 3Train Naive Bayes
  4. 4Evaluate with a confusion matrix
  5. 5Test on your own messages

Go further

  • + Compare with logistic regression
  • + Explain which words drive predictions

Prepare with

Deep Learning

Sentiment Analysis

Decide whether a review is positive or negative with a neural network.

  • Tokenisation
  • Embeddings
  • Neural networks
Milestones and next steps
  1. 1Prepare review text
  2. 2Build a vocabulary
  3. 3Train an embedding + LSTM model
  4. 4Evaluate accuracy
  5. 5Try it on new reviews

Go further

  • + Swap in a pretrained Transformer
  • + Deploy as a small web app

Prepare with

Machine Learning

Customer Churn Prediction

Find customers likely to leave and explain the main reasons.

  • Classification
  • Imbalanced data
  • Explainability
Milestones and next steps
  1. 1Explore churn rates
  2. 2Encode and scale features
  3. 3Train and compare models
  4. 4Handle class imbalance
  5. 5Explain top churn drivers

Go further

  • + Estimate the money saved by a retention campaign
  • + Build a monitoring report

Prepare with

Machine Learning

Customer Segmentation

Group customers by behaviour with clustering and describe each segment.

  • Clustering
  • Scaling
  • Visualisation
Milestones and next steps
  1. 1Create behaviour features
  2. 2Scale the data
  3. 3Choose K with the elbow method
  4. 4Fit K-Means
  5. 5Profile and name each segment

Go further

  • + Compare with DBSCAN
  • + Visualise segments with PCA

Prepare with

Machine Learning

Sales Forecasting

Forecast next month's sales from historical data.

  • Time series
  • Regression
  • Validation
Milestones and next steps
  1. 1Plot trend and seasonality
  2. 2Create lag features
  3. 3Split by time
  4. 4Train a forecasting model
  5. 5Measure forecast error

Go further

  • + Add holiday effects
  • + Forecast multiple products

Prepare with

Deep Learning

Handwritten Digit Recognition

Train a neural network that reads handwritten digits from the MNIST dataset.

  • Neural networks
  • Training loops
  • Evaluation
Milestones and next steps
  1. 1Load MNIST
  2. 2Build a small network
  3. 3Write the training loop
  4. 4Reach 97% accuracy
  5. 5Inspect the mistakes

Go further

  • + Switch to a CNN
  • + Draw digits in the browser and classify them

Prepare with

Computer Vision

Image Classification

Fine-tune a pretrained CNN to recognise your own image categories.

  • CNNs
  • Transfer learning
  • Augmentation
Milestones and next steps
  1. 1Collect and split images
  2. 2Load a pretrained model
  3. 3Replace the final layer
  4. 4Fine-tune with augmentation
  5. 5Evaluate per class

Go further

  • + Export to ONNX
  • + Build a drag-and-drop demo

Prepare with

Computer Vision

Face Detection

Detect faces in photos and a live webcam feed with OpenCV.

  • OpenCV
  • Detection
  • Video
Milestones and next steps
  1. 1Read images and video
  2. 2Load a face detector
  3. 3Draw boxes on faces
  4. 4Run on a webcam stream
  5. 5Measure frames per second

Go further

  • + Blur faces for privacy
  • + Count people over time

Prepare with

Computer Vision

Document Scanner

Turn a phone photo of a page into a flat, clean scan.

  • Edges
  • Contours
  • Perspective transform
Milestones and next steps
  1. 1Detect page edges
  2. 2Find the page contour
  3. 3Apply a perspective warp
  4. 4Threshold for a clean scan
  5. 5Save as PDF

Go further

  • + Add OCR to extract text
  • + Handle multiple pages

Prepare with

Generative AI

PDF Question Answering

Upload a PDF and ask questions that are answered with page citations.

  • Chunking
  • Embeddings
  • RAG
Milestones and next steps
  1. 1Extract text from the PDF
  2. 2Split into chunks
  3. 3Embed and index chunks
  4. 4Retrieve for a question
  5. 5Generate an answer with citations

Go further

  • + Support multiple files
  • + Add reranking and evaluate quality

Prepare with

Generative AI

Document Summarizer

Summarise long documents reliably, even beyond the context window.

  • Prompting
  • Chunking
  • Map-reduce
Milestones and next steps
  1. 1Summarise a short text
  2. 2Split long documents
  3. 3Summarise chunks then combine
  4. 4Control length and tone
  5. 5Check for missed facts

Go further

  • + Summarise in multiple languages
  • + Output structured key points

Prepare with

Generative AI

Website AI Assistant

A chat widget that answers questions about a website's own content.

  • Crawling
  • RAG
  • Streaming UI
Milestones and next steps
  1. 1Collect site pages
  2. 2Index the content
  3. 3Build the answer API
  4. 4Add a streaming chat widget
  5. 5Refuse off-topic questions

Go further

  • + Show source links
  • + Log unanswered questions

Prepare with

AI Agents

Personal AI Assistant

An agent that manages notes and to-dos and looks things up using tools.

  • Tool calling
  • Agent loop
  • State
Milestones and next steps
  1. 1Define note and to-do tools
  2. 2Implement the agent loop
  3. 3Add a search tool
  4. 4Persist data between runs
  5. 5Confirm before destructive actions

Go further

  • + Add long-term memory
  • + Connect a calendar API

Prepare with

Advanced projects

Production-style systems that mirror real engineering work.

AI Agents

Research Agent

An agent that plans a search, reads sources and writes a cited report.

  • Planning
  • Search tools
  • LangGraph
Milestones and next steps
  1. 1Plan sub-questions
  2. 2Search and fetch sources
  3. 3Extract key facts
  4. 4Write a cited report
  5. 5Add a self-review step

Go further

  • + Run sub-questions in parallel
  • + Add human approval of the plan

Prepare with

AI Agents

Customer Support Agent

An agent that answers from a knowledge base, looks up orders and hands off to a human.

  • RAG
  • Tools
  • Guardrails
Milestones and next steps
  1. 1Answer FAQs with RAG
  2. 2Add an order-lookup tool
  3. 3Route by intent
  4. 4Escalate to a human
  5. 5Add input and output guardrails

Go further

  • + Evaluate with a test set of tickets
  • + Track resolution rate

Prepare with

AI Agents

Multi-Agent Research System

A team of agents (planner, researchers and writer) coordinated by an orchestrator.

  • Orchestration
  • Handoffs
  • Shared state
Milestones and next steps
  1. 1Design agent roles
  2. 2Build the orchestrator
  3. 3Share state between agents
  4. 4Merge findings
  5. 5Trace and measure cost

Go further

  • + Add a critic agent
  • + Compare against a single agent

Prepare with

Generative AI

AI Knowledge Base

A searchable company knowledge base with ingestion, permissions and evaluation.

  • Ingestion pipelines
  • Hybrid search
  • Evaluation
Milestones and next steps
  1. 1Ingest many file types
  2. 2Add metadata filters
  3. 3Combine keyword and vector search
  4. 4Answer with citations
  5. 5Build an evaluation set

Go further

  • + Add per-user permissions
  • + Schedule re-indexing

Prepare with

Generative AI

Multimodal Assistant

An assistant that understands text, images and voice in one conversation.

  • Vision input
  • Speech
  • Tool calling
Milestones and next steps
  1. 1Accept image uploads
  2. 2Answer questions about images
  3. 3Add speech-to-text
  4. 4Reply with text-to-speech
  5. 5Keep a mixed conversation history

Go further

  • + Read documents and charts
  • + Add real-time voice

Prepare with

Computer Vision

Real-Time Object Recognition

Detect objects in live video with a YOLO model trained on your own classes.

  • YOLO
  • Custom datasets
  • Real-time inference
Milestones and next steps
  1. 1Label a small dataset
  2. 2Train YOLO
  3. 3Validate with mAP
  4. 4Run on live video
  5. 5Optimise for speed

Go further

  • + Track objects across frames
  • + Deploy to an edge device

Prepare with

MLOps & AI Engineering

Production AI API

A hardened API around an AI model with auth, rate limits, tests and containers.

  • FastAPI
  • Docker
  • Testing
Milestones and next steps
  1. 1Design the endpoints
  2. 2Add validation and auth
  3. 3Add rate limiting and logging
  4. 4Write tests
  5. 5Containerise and deploy

Go further

  • + Add streaming
  • + Add tracing and a cost dashboard

Prepare with

MLOps & AI Engineering

End-to-End ML Deployment

Take a model from training to a monitored service with an automated pipeline.

  • MLflow
  • CI/CD
  • Monitoring
Milestones and next steps
  1. 1Track training with MLflow
  2. 2Register the best model
  3. 3Serve it in a container
  4. 4Automate with CI/CD
  5. 5Monitor drift and latency

Go further

  • + Add canary releases
  • + Trigger automatic retraining

Prepare with