Career Program · 3rd & Final Year B.Tech
AI Engineering
& Agentic Systems
Build, evaluate and deploy real AI applications — and walk into AI Engineer and GenAI interviews with a portfolio to prove it.
- Weeks
- 12
- Live sessions
- 24
- Hours of hands-on instruction
- 48
- Portfolio projects
- 3
Weeks
Live sessions
Hours of hands-on instruction
Portfolio projects
- 2 live sessions per week, 2 hours each
- Build-along format: you build in every session
- 3rd & final year B.Tech students
- Certificate of completion
Why this program
From “knows the theory” to “can ship something that works”
AI hiring has moved to proof of capability. You start by building visible, working products with the tools professionals use today (Claude, ChatGPT, Cursor, workflow automation platforms), then learn the engineering underneath: Python, APIs, data, machine learning, LLMs, retrieval, tools and MCP, and agents. You finish by deploying a capstone system and practising for interviews.
How you learn: see it work → build it → break it → understand why → improve it → measure it. Theory arrives exactly when it helps you make a better engineering decision.
What you will be able to do
- Use Claude, ChatGPT and Cursor to research, design, build, test and document, while verifying what AI produces rather than trusting it blindly.
- Automate real workflows (email, documents, forms, data entry, reports) with AI-powered automation platforms.
- Build full-stack AI applications with a UI, API, database, validation and tests.
- Train and evaluate classical ML models and serve them through an API.
- Explain how LLMs work (tokens, attention, context, decoding) and choose between cloud and local models, prompting, RAG and fine-tuning.
- Build tool-using assistants and MCP servers with proper permissions and guardrails.
- Build and evaluate RAG systems with hybrid retrieval, reranking, citations and access control.
- Design agents and multi-agent workflows with state, human approval, safety limits and tracing.
- Containerise, test, deploy and monitor an AI service.
- Scope a problem like a customer-facing engineer and defend your design in technical interviews.
Careers
Industry roles this program prepares you for
Each role maps to concrete evidence you will have built by the end of the program.
AI Engineer
Builds and runs end-to-end AI systems in production
You will show: Capstone system with deployment, tracing, evaluation and runbook
GenAI Engineer
Builds LLM applications using prompting, structured outputs, RAG, tools and guardrails
You will show: Project 2 (RAG) with evaluation set, plus the tool-using assistant and MCP server
Applied AI Developer
Integrates AI into full-stack products
You will show: Project 1: UI, API, database, validation and tests
Junior ML Engineer
Prepares data, trains and evaluates models, serves them via APIs
You will show: ML mini-project with pipeline, MLflow tracking, model card and API
AI Automation Engineer
Automates business workflows with AI, connectors and approvals
You will show: Workflow automations, long-running workflows and human-in-the-loop agents
AI Solutions Engineer
Translates customer needs into AI solutions and demos
You will show: Discovery notes, scope, architecture and demo from the scoping session and capstone
Forward Deployed Engineer (entry level)
Works directly with customers to scope, build, deploy and hand off AI systems
You will show: Customer discovery and scoping, production deployment, runbook, change-request handling and a stakeholder-ready demo
Forward Deployed Engineer roles at many companies ask for prior industry experience. This program builds the core foundations (discovery, scoping, hands-on delivery, deployment and handoff) that make you a credible candidate for associate and entry-level FDE openings.
Program structure
Twelve weeks, three phases, one portfolio
Two live sessions a week on weekdays, plus weekly exercises you complete on your own schedule.
Phase 1 — Build with AI
AI tools, automation, Python, APIs and ML foundations.
Milestone: Deployed AI app + ML model API
Phase 2 — GenAI Engineering
LLM internals, tools, MCP and retrieval-augmented generation.
Milestone: Evaluated RAG system
Phase 3 — Agents, Production & Placement
Agents, deployment, system design and interview readiness.
Milestone: Capstone + interview assessment
Every session (2 hours)
- 10 minRecap and a live demo of what you will build
- 25 minCore concepts and architecture
- 70 minBuild-along: you build with the instructor, step by step
- 15 minWhat went wrong, what to improve, and the weekly exercise
Before you join
- Basic programming knowledge in any language (C, C++, Java or Python) and familiarity with OOP concepts.
- A laptop (8 GB RAM minimum, 16 GB recommended) and a stable internet connection.
- Free accounts on GitHub and the AI platforms we use (setup guide shared before the first session).
- Curiosity and willingness to build things that fail before they work.
Detailed curriculum
Week by week, session by session
Open any week to see what you will build and the concepts covered.
Weeks 1–4
Phase 1 — Build with AI
1Week 1 · Sessions 1 & 2Your AI Toolkit: Build Something in Session One
Session 1 — Claude and ChatGPT as Power Tools
You will build: A working web app (calculator, dashboard or study planner) from a plain-English description, in under an hour using Claude Artifacts.
- Models vs products vs workflows vs agents: how the AI landscape fits together
- Claude Projects, files, artifacts and ChatGPT data analysis, files and browsing
- Prompt and context engineering: goal, context, constraints, examples, output format
- Using AI to analyse a dataset and a document, then produce a sourced decision brief
- Multimodal work: reading documents, tables, images and audio
Session 2 — Cursor: From Idea to Running Application
You will build: A complete multi-page website or tool with Cursor, on your own machine, running locally.
- Cursor interface, chat vs agent mode, codebase questions, planning and editing
- Writing requirements as prompts; iterating in small, reviewable steps
- Verification and responsible use: hallucinations, insecure defaults, fake APIs, copyright, data privacy
- AI-use log: how professionals document and review AI-generated work
Week 1 practice: Build a personal portfolio page with Cursor and write a one-page verification report on what the AI got wrong. Quiz: AI fundamentals and prompting.
2Week 2 · Sessions 3 & 4Automate Workflows and Ship Code
Session 3 — AI Workflow Automation
You will build: An end-to-end automation: incoming email or form → AI classification and summary → spreadsheet/CRM entry → notification.
- Automation platforms (n8n and similar), triggers, webhooks and connectors
- Putting an LLM inside a workflow: extraction, classification, drafting
- Human-in-the-loop approvals, error handling and logging
- Document and invoice processing with OCR and multimodal models
Session 4 — Coding Agents, Git and Deployment
You will build: Push your Week 1 app to GitHub, add a feature through a pull request using a coding agent, and deploy it to a live URL.
- Git and GitHub essentials: commits, branches, pull requests
- Claude Code and Cursor agents; rules files (AGENTS.md, CLAUDE.md) to keep agents on-task
- Reviewing AI-generated diffs; small-change discipline
- One-click deployment of a web app
Week 2 practice: Automate one real task from your own life (placement mails, notes, expense logs) and deploy your portfolio page. Quiz: Automation and Git.
3Week 3 · Sessions 5 & 6Python, APIs and Structured AI
Session 5 — Python Essentials for AI Builders
You will build: A command-line tool that reads files, calls an API and writes structured output.
- Functions, modules, classes, type hints, exceptions, logging, virtual environments
- Reading and debugging AI-generated Python
- Files, JSON, pandas basics, async programming and concurrency limits
- Testing with pytest
Session 6 — APIs, FastAPI and Structured Outputs
You will build: A FastAPI service that wraps an LLM, returns validated JSON and handles failures gracefully.
- HTTP, status codes, API keys and secrets, rate limits
- Calling model APIs; schema-constrained outputs with Pydantic; validation, retries and repair
- Splitting a prompt into extraction → classification → generation steps
- Building an evaluation set of 20 test cases
Week 3 practice: Build a resume-parser or invoice-extractor API and measure its accuracy on 20 cases. Quiz: Python and APIs.
4Week 4 · Sessions 7 & 8Data, Machine Learning and Project 1
Session 7 — SQL, Data Persistence and Your First AI App
You will build: A Streamlit/Gradio (or React) app with login, a database and stored history, powered by your Week 3 API.
- SQL: SELECT, joins, aggregation, indexes, transactions; parameterised queries and SQL injection
- SQLite/PostgreSQL with SQLModel; storing users, jobs and outputs
- Interface basics: file upload, streaming, loading and error states
- Audit records and data deletion
Session 8 — Machine Learning in Practice
You will build: An end-to-end ML model: data cleaning → pipeline → evaluation → API.
- Data wrangling, missing values, leakage and train/validation/test splits
- Linear and logistic regression, decision trees, Random Forest and gradient boosting
- Bias–variance, overfitting, regularisation, cross-validation
- Precision, recall, F1, ROC-AUC, threshold selection and error analysis
- scikit-learn pipelines, MLflow tracking, model cards, serving with FastAPI
Week 4 practice: Project 1 (due end of Week 4): a deployed AI application with UI, API, database, validation and at least 20 tests, plus an ML mini-project with a model API and model card. Evaluation 1: project review, ML and Python/SQL online assessment.
Weeks 5–8
Phase 2 — GenAI Engineering
5Week 5 · Sessions 9 & 10Inside LLMs and Choosing Models
Session 9 opens with a showcase of Project 1.
Session 9 — Neural Networks and Transformers
You will build: A small neural network in PyTorch, then explore tokenisation and attention with pretrained models.
- Layers, activations, loss, backpropagation intuition, optimisers, overfitting, dropout
- Tokens, embeddings, attention, positional information, context windows, KV cache
- Encoder, decoder and encoder-decoder models; BERT-style vs GPT-style
- Decoding: temperature, top-p, deterministic evaluation
Session 10 — Hugging Face, Local Models and Model Selection
You will build: A cloud vs local model comparison on a fixed test set, with a routing policy.
- Hugging Face models, datasets, pipelines, model cards and licences
- Running local models with Ollama; quantisation and hardware trade-offs
- Routing tasks by sensitivity, quality, latency and cost
- When to use prompting, RAG or fine-tuning (PEFT and LoRA overview)
Week 5 practice: Model comparison report with a recommendation. Quiz: Transformers and LLM concepts.
6Week 6 · Sessions 11 & 12Tools, Guardrails and MCP
Session 11 — Function Calling and Safe Tool Design
You will build: An assistant that uses three tools (search, lookup, draft creation) with logging and a step budget.
- Function calling: tool descriptions, typed inputs, read-only vs mutating tools
- Idempotency, timeouts, retries and error handling
- Least privilege, allowlists, human approval for sensitive actions
- Prompt injection and the basics of threat modelling
Session 12 — Model Context Protocol (MCP)
You will build: Your own MCP server, connected to Claude or Cursor.
- MCP hosts, clients, servers, tools, resources and prompts
- Capability discovery, transports, authentication
- OAuth concepts, consent, token handling and connector security
- Reviewing an unsafe connector and hardening it
Week 6 practice: Evaluate your assistant on 25 tasks for tool selection, argument correctness and unauthorised actions. Quiz: Tools, MCP and security.
7Week 7 · Sessions 13 & 14Retrieval-Augmented Generation (RAG)
Session 13 — Embeddings, Vector Search and Ingestion
You will build: A document ingestion pipeline and a searchable vector index.
- Embeddings, cosine similarity, ANN indexes, top-k retrieval
- Chroma/Qdrant; metadata filters and access-control filters
- Parsing, cleaning, chunk size, overlap, structure-aware chunking, tables
- Deduplication, incremental ingestion and versioning
Session 14 — Hybrid Retrieval, Reranking and Grounded Answers
You will build: A RAG chatbot with citations and the ability to say “I don't know.”
- Keyword + vector hybrid search, rerankers, context assembly
- Citations, abstention and grounding
- Prompt injection through retrieved content; per-user access filtering
- Query rewriting and multi-step retrieval
Week 7 practice: Compare three chunking strategies and report retrieval changes. Quiz: Retrieval and RAG design.
8Week 8 · Sessions 15 & 16RAG Evaluation and Customer Scoping
Session 15 — Evaluating and Observing RAG Systems
You will build: An evaluation dataset, a baseline score and a trace-based debugging workflow.
- Retrieval relevance, context recall, groundedness, answer correctness, abstention
- Human scoring, rule-based checks and calibrated LLM-as-judge
- Tracing and regression testing with LangSmith, Langfuse or OpenTelemetry
- Cost and latency tracking
Session 16 — Customer Discovery and Solution Scoping
You will build: A simulated customer interview, then a scope, success metric and acceptance criteria for your capstone.
- Stakeholder interviews, current-state workflow mapping, pain points
- Proof of concept vs pilot vs production; non-goals and risks
- Statement of work, delivery plan and executive updates
- Translating business needs into architecture
Week 8 practice: Project 2 (due end of Week 8, teams of 3–4): a domain RAG system with hybrid retrieval, citations, access control, an evaluation set of at least 30 cases and a regression report. Evaluation 2: project review and RAG online assessment. Capstone topics are selected.
Weeks 9–12
Phase 3 — Agents, Production & Placement
9Week 9 · Sessions 17 & 18Agent Foundations and Orchestration
Session 17 opens with a showcase of Project 2.
Session 17 — Workflows, Agents and Memory
You will build: The same task as a fixed workflow and as a bounded agent, then compare them.
- Workflow vs agent vs simple function: when autonomy helps and when it hurts
- The reasoning-and-tool loop, termination and step budgets
- Working memory, durable state, long-term memory; tenant isolation and deletion
Session 18 — Graph Orchestration and Human Approval
You will build: A stateful agent graph that pauses for approval, survives a restart and resumes.
- Nodes, edges, routers, typed state, conditional and parallel branches (LangGraph)
- Checkpoints, retries, idempotency and cancellation
- Approval packets: context, proposed action and impact
Week 9 practice: Capstone architecture and repository setup. Quiz: Agents and orchestration.
10Week 10 · Sessions 19 & 20Multi-Agent Systems, Safety and Evaluation
Session 19 — Multi-Agent Patterns and Agent Safety
You will build: A router or supervisor-worker system, compared against a single-agent baseline.
- Handoffs, supervisor-worker, planner-executor, generator-critic patterns
- Bounded iteration and correlated model errors
- Red-teaming agents: injection, exfiltration, unsafe delegation; spend limits and kill switches
Session 20 — Agent Evaluation, Observability and Long-Running Workflows
You will build: Traces, a regression suite and a long-running workflow with status tracking.
- Traces, spans, trajectory evaluation, tool correctness, policy compliance
- Offline test suites and production sampling
- Webhooks, queues, workers, retries and dead-letter handling; n8n for low-code orchestration
Week 10 practice: Capstone vertical slice with tracing and a 30-case evaluation set. Evaluation 3: mid-capstone architecture and safety review.
11Week 11 · Sessions 21 & 22Production Engineering
Session 21 — Containers, CI/CD and Testing AI Systems
You will build: A Dockerised capstone with a CI pipeline that blocks failing tests, broken schemas and leaked secrets.
- Docker and Docker Compose; configuration, health checks, dependency pinning
- Unit, integration, contract and end-to-end tests; mocking model calls
- GitHub Actions; evaluations as regression gates
Session 22 — Deployment, Cost and Operations
You will build: A deployed capstone with monitoring, a load test and a runbook.
- Cloud deployment options, rate limits, caching, batching, model fallback
- Latency and cost optimisation; awareness of vLLM and Triton serving
- Monitoring, incident response, rollback and handoff documentation
- Handling change requests from stakeholders
Week 11 practice: Complete capstone deployment, runbook and demo script. Quiz: Deployment and operations.
12Week 12 · Sessions 23 & 24Capstone and Interview Readiness
Session 23 — AI System Design and Interview Intensive
You will build: Design and defend an AI system on a whiteboard.
- A repeatable framework: requirements → data → retrieval/agents → evaluation → safety → deployment → cost
- Worked designs: enterprise document Q&A, customer-support agent, LLM gateway
- Technical interview patterns for Python, SQL, ML, LLM, RAG and agent questions
- Resume, GitHub profile and project write-ups; behavioural answers using the STAR method
Session 24 — Capstone Demo Day and Mock Interviews
You will build: Present your capstone and take part in a mock interview round.
- 10-minute capstone demo, architecture walkthrough and panel questions
- Individual technical and system-design mock interview with feedback
Week 12 practice: Evaluation 4: capstone evaluation and final interview assessment.
Portfolio
Three projects you can put in front of recruiters
Every capstone repository ships with a README, architecture diagram, decision records, evaluation report, security notes, runbook, demo video and AI-use log.
Project 1 — AI Application + ML Mini-Project
Weeks 3–4 · Individual
Deployed app with UI, API, database, validation and 20+ tests; ML pipeline with model card and API.
Project 2 — Domain RAG System
Weeks 7–8 · Teams of 3–4
Hybrid retrieval, citations, access control, 30+ case evaluation and a regression report.
Capstone — Agentic AI System
Weeks 9–12 · Teams of 3–4
Stateful agent workflow with tools/MCP, approvals, safety controls, tracing, evaluation suite, Docker + CI, runbook and demo.
Capstone ideas
- Enterprise knowledge copilot
- Support-ticket triage and resolution agent
- HR screening assistant
- Invoice and document automation
- Campus helpdesk assistant
- Sales research and outreach drafter with approval
Placement readiness
Interview preparation, built into every week
Each week includes a short, self-paced practice set. By Week 12 you will have worked through roughly 40 coding problems and a complete question bank.
| Week | Concept questions | Coding practice |
|---|---|---|
| 1 | LLM basics, prompting, hallucinationPython warm-ups | Python warm-ups |
| 2 | Git, automation, AI-assisted developmentArrays, strings, hashing | Arrays, strings, hashing |
| 3 | REST, async, testing, structured outputsSQL queries, sliding window | SQL queries, sliding window |
| 4 | ML basics, bias–variance, metrics, leakageStacks, queues, linked lists | Stacks, queues, linked lists |
| 5 | Transformers, attention, tokenisation, KV cacheTrees, recursion | Trees, recursion |
| 6 | Tool calling, MCP, prompt injectionBinary search, heaps | Binary search, heaps |
| 7 | RAG design and failure modesGraphs: BFS/DFS | Graphs: BFS/DFS |
| 8 | RAG evaluationDynamic programming basics | Dynamic programming basics |
| 9 | Agents, memory, orchestrationMixed medium problems | Mixed medium problems |
| 10 | Multi-agent systems, safety, evaluationMixed medium problems | Mixed medium problems |
| 11 | Deployment, scaling, costTimed practice set | Timed practice set |
| 12 | Full revision, system design, behaviouralTimed mock test + mock interview | Timed mock test + mock interview |
Assessment & certification
Evaluated on what you build
Weekly quizzes, two milestone project evaluations, a capstone evaluation and a final interview assessment.
How you are scored
- Weekly exercises and quizzes20%
- Project 1 + ML mini-project15%
- Project 2 (RAG)20%
- Capstone30%
- Final interview assessment10%
- Participation5%
Certification
Certificate of Completion
70% overall, attendance in at least 20 of 24 sessions, and submission of all three projects.
Certificate with Distinction
85% overall and a strong final interview assessment.
What you leave with
- A GitHub portfolio with three substantial, deployed projects
- Evaluation reports and architecture write-ups you can discuss in interviews
- A polished resume, project write-ups and rehearsed behavioural stories
- Hands-on experience with the stack hiring teams ask about: LLM APIs, RAG, MCP, agents, FastAPI, Docker, CI/CD and observability
The program builds skills and evidence of capability. It does not guarantee placement; outcomes depend on your effort, projects and interview performance.
Toolkit
Tools and technologies you will use
Most tools offer free tiers. Tools evolve quickly, so the curriculum focuses on concepts that transfer across platforms.
- Claude
- ChatGPT
- Cursor
- Claude Code
- n8n
- Python
- FastAPI
- SQL
- scikit-learn
- PyTorch
- Hugging Face
- Ollama
- Chroma / Qdrant
- MCP
- LangGraph
- Docker
- GitHub Actions
- LangSmith / Langfuse
FAQ
Questions students ask
No. You need basic programming ability; Python is covered in Week 3.
Ready to build your AI career?
Register your interest and we'll share upcoming batch dates and fees, or talk to us if you're a college or training partner looking to bring this program to your students.