The learning journey
Six topics, one path, start to finish
The journey starts at securing a web application and ends with understanding the model inside an LLM. As you scroll, the line fills and each topic lights up in turn. Every topic comes with interactive lessons on PurpleLane LMS, ends with a quiz, and submits its work as a Git check-in.
1
1Authentication & AuthorizationWeek 1
How a real website knows who you are and what you may do. Students secure ShopMATE end to end: hashed passwords, sessions and JWT tokens, user and admin roles, and security best practices, testing each endpoint with Postman. Google sign-in is an optional extension.
ExpressJWTMongoDBPostmanOAuth (optional)
After this topicA student can build a secure login and signup system with role based access, and keep admin-only pages protected from regular users.
2
2Best Practices for Managing Codebase with Git & GitHubWeek 1 · 1 day
A short, practical session: repositories, commits, branches, pull requests, and resolving merge conflicts. From this day on, every task in the internship is submitted as a Git check-in.
GitGitHubPull RequestsMerge Conflicts
After this topicA student can manage a codebase with Git: branch, commit, open pull requests, and resolve merge conflicts when two changes collide.
3
3Agentic AI & LangChainWeek 2
From a chatbot that only talks to an agent that takes action: tools, function calling, and the ReAct loop, built with LangChain.js on each student's own free Gemini API key.
LangChain.jsGemini 2.5 Flash
After this topicA student can build an AI agent that uses tools to take real actions in an application, and send each user request to the right handler.
4
4Multi-Agent OrchestrationWeek 3
Several agents working together with LangGraph: router, pipeline, and supervisor patterns built as state graphs, with a human approval step before anything is final.
LangGraph.jsState GraphsHuman in the Loop
After this topicA student can coordinate several agents in one workflow, add an async human approval step, and track the state of each item in the database and the browser.
5
5LLMOpsWeek 3
Running AI features responsibly: observability and logging, tracing with LangSmith, guardrails, evaluation, and cost control.
LangSmithGuardrailsCost & Tokens
After this topicA student can run AI features responsibly in production: trace each request, track cost and logs, and defend an agent against harmful or off-topic input.
6
6Foundations of Deep LearningWeek 4
The closing topic opens the model itself. Interactive modules build from a single neuron to the Transformer inside an LLM: neurons and training, CNNs, embeddings, attention and Transformers, and how an LLM is made. RNNs are an optional module, covered if time permits. No heavy mathematics.
Neural NetworksEmbeddingsAttention & Transformers
After this topicA student can explain how a neural network learns, build and train one on real data, and read the architecture of a modern LLM with confidence.
One project runs through everything
ShopMATE grows with the students for all 4 weeks
Students build on ShopMATE, a MERN e-commerce app (MongoDB, Express, React, Node.js). Across the program they secure it, give it AI agents, run those agents responsibly, and then open the model to see how it works inside. They always know why each topic matters, because they ship it into the same app.
Students do not start from a blank screen. We hand them a working ShopMATE boilerplate codebase, the storefront, server, and database already in place, so all of their effort goes into building the GenAI features on top of it, not into plumbing.
Foundations
Week 1 · Auth, Git, Postman
Agents
Week 2 · Agentic AI & LangChain
Multi-Agent & Ops
Week 3 · LangGraph, LLMOps
Deep Learning
Week 4 · Neuron to Transformer
Program at a Glance
| Week | Topics | What Students Build | Assessment |
|---|---|---|---|
| Week 1 | Authentication & Authorization · Git & GitHub · Postman | Secure login, JWT, roles, optional Google sign-in on ShopMATE, with endpoints tested in Postman; the repository every later task is checked into | Auth quiz and a 15-question Git quiz on PurpleLane LMS · first Git check-ins |
| Week 2 | Agentic AI & LangChain | A query-routing agent and an AI assistant that updates a product's stock and price, on each student's own free Gemini key | Topic quiz on PurpleLane LMS · two task check-ins |
| Week 3 | Multi-Agent Orchestration · LLMOps | A multi-agent blog pipeline with human approval stored in the database, a cost and logs dashboard, and a guardrail layer | Multi-Agent and LLMOps quizzes on PurpleLane LMS · three task check-ins |
| Week 4 | Foundations of Deep Learning | Interactive modules from a single neuron to the Transformer, plus three builds: a first neural network, MNIST, and a Fashion CNN | Topic quiz on PurpleLane LMS · three task check-ins · final check-in of the complete project |
Week 1Secure Foundations Foundations · Auth, Git & Postman
Topic 1Authentication & AuthorizationLessons + Tasks
ExpressJWTMongoDBPostmanOAuth 2.0
Who are you, and what may you do?
Authentication answers the first question, authorization the second. How login works: passwords hashed and salted, never stored as plain text.
Sessions, JWT tokens, roles & permissions
How a server remembers a logged-in user, and how user and admin roles protect routes. Students add both to ShopMATE.
Test your endpoints with Postman
Send real requests to each route, check the response and status code, and confirm that protected routes reject a request with no valid token.
"Login with Google" (optional)
How third-party login works: the redirect, the consent screen, the token exchange. An optional extension for students who finish early.
Security best practices
Rate limiting, HTTPS, secret management, and the most common attacks, with ShopMATE hardened against each one.
Topic 2 · 1 DayManaging Codebase with Git & GitHubHands-On
GitGitHubVS Code
Repositories, commits, branches
A compact, practical session: track every change, write clear commit messages, branch for new work, merge when ready.
GitHub: push, pull, pull requests
Each student publishes their own ShopMATE repository and learns the pull request flow used in real engineering teams.
Resolving merge conflicts
What a merge conflict is, why it happens when two changes touch the same lines, and how to read the markers and fix it calmly.
Check-ins begin here
From this day onward, every task in the internship is submitted as a Git check-in. By the end, the commit history tells the whole story.
Auth quiz and a 15-question Git quiz on PurpleLane LMS
First Git check-ins of secured ShopMATE
Week 2Agentic AI & LangChain Agents · Topic 3
Topic 3Agentic AI & LangChainConcepts + Build
LangChain.jsGemini 2.5 Flash
From chatbot to agent
A chatbot only talks. An agent decides, uses tools, and acts. What changes, and why it matters for a real store.
Tools & function calling
How an LLM asks the application to run a function, with typed inputs, and gets the result back to continue reasoning.
The ReAct loop, then build it
Reason, act, observe, repeat. First the idea, then a working agent in ShopMATE, on each student's own free Gemini API key.
Types of agents
From a simple tool-caller to planning agents. When a plain chatbot is enough and when an agent is worth the cost.
Task 1 · Route customer queries
An agent that reads each customer query and sends it to the right handler, built with LangChain.
Task 2 · Stock and price assistant
An AI assistant that can update a product's stock, its price, or both, by calling tools on the ShopMATE backend.
Topic quiz on PurpleLane LMS
Two task check-ins (routing agent, stock/price assistant)
Week 3Multi-Agent Orchestration & LLMOps Orchestration & Ops · Topics 4 & 5
Topic 4Multi-Agent OrchestrationConcepts + Tasks
LangGraph.jsLangChain.jsGemini 2.5 Flash
State graphs: nodes, edges, state
LangGraph turns an agent workflow into a graph the team can read, test, and extend. The three core ideas, step by step.
Router, pipeline, supervisor
The three orchestration patterns, and how to pick the right one for a job. Built, not just described.
Task · Blog generator with human approval
Several agents coordinate to draft a post, then the pipeline pauses for a person to approve it before publishing. All generated content is stored in the database, and the browser shows each item's state: draft, approved, rejected, or published.
Topic 5LLMOps · Running AI ResponsiblyConcepts + Tasks
LangSmithMongoDBExpressGemini 2.5 Flash
Observability, logging & tracing
Structured logs for every model call, and full request traces in LangSmith, so nothing the agent does is a mystery.
Guardrails & evaluation
Checks before and after the model: decline off-topic input, stop bad output, and score answer quality on a test set.
Cost & token economics
Why output tokens cost more than input, what real incidents have cost companies, and the four levers that control spend.
Task 1 · Cost and logs dashboard
Usage and request logs recorded in MongoDB, shown on an admin dashboard protected by the Week 1 role check.
Task 2 · Guardrail layer on the agents
A guardrail layer in front of the AI agents built earlier, resisting prompt injection. Defence in depth, tested with real attack attempts.
Multi-Agent quiz on PurpleLane LMS
LLMOps quiz on PurpleLane LMS
Three task check-ins
Week 4 Foundations of Deep LearningTopic 6 · The Closing Topic
Half AFrom a Neuron to a NetworkInteractive Modules
Visual DemosNo Heavy Math
Neurons, layers, and learning
What a neuron computes, how layers stack, and how loss and gradient descent slowly make a network better.
Training in practice
Overfitting, validation, and the habits that separate a model that memorizes from a model that generalizes.
CNNs: networks that see
How a small filter slides over an image to find patterns, and why reusing it makes vision practical.
RNNs: networks that read in order (optional)
Why word order matters, how a hidden state carries memory through a sentence, and where RNNs reach their limits. Covered if time permits.
Half BFrom Embeddings to the LLMInteractive Modules
Visual DemosTies to Week 2
Embeddings from the inside
The vectors students have used since their RAG days, now explained from the inside: meaning as position in space.
Attention & Transformers
The idea that replaced RNNs: every word looks at every other word at once. The architecture behind every modern LLM.
From Transformer to LLM
Pre-training, instruction tuning, and why Gemini can follow instructions at all. The Week 2 magic, demystified.
Transfer learning & fine-tuning, in concept
Why nobody trains from zero: start from a model that already knows language, then teach it your task. The idea, explained clearly.
Hands-OnThree Builds, From a Neuron to a CNNTask Check-ins
Neural NetworksMNISTFashion Dataset
Task 1 · Your first neural network
Build a small artificial neural network from the ground up, and watch it learn a simple pattern.
Task 2 · MNIST digits
Train a network to read handwritten digits on the MNIST dataset, using gradient descent or Adam.
Task 3 · Fashion images with a CNN
Classify clothing images from the Fashion dataset with a convolutional neural network, the network type that sees.
Technologies You'll Work With
Language
JavaScript
Runtime
Node.js
Backend
Express
Database
MongoDB
Frontend
React
Version Control
Git
Check-ins
GitHub
JWT
Auth
JWT + OAuth 2.0
API Testing
Postman
AI Framework
LangChain.js
AI Framework
LangGraph.js
AI Framework
LangSmith.js
LLM
Gemini 2.5 Flash
All tools above are free for students.