Skip to content
KKOMPAI
Sign inApply now
LLMsRAGAgentsMCPVoice AIDeployment

The AI Engineering Masterclass

Eight weekends of building. Every session ends with something that runs.

K
Karthik
Lead AI/ML Engineer (MLOps), UPS

About this program

Most AI courses hand you concepts and hope you build later. This one inverts it: the concept exists to serve the thing you ship that same afternoon. You finish with a portfolio, not a certificate — your repos, your domains, your API keys.

What you'll ship

  • ✓Week 1. A custom Claude Code skill that drafts technical posts from your week's code activity, plus the reasoning engine behind an AI travel planner.
  • ✓Week 2. A document assistant that answers from internal policy PDFs with page-level citations, and a knowledge graph for multi-hop questions.
  • ✓Week 3. A live portfolio website, and the Week 1 travel planner wired to a React front end on a public URL.
  • ✓Week 4. Your own MCP server with authenticated access to your machine, plus an agent that tailors a PDF resume per job listing.
  • ✓Week 5. Two voice agents: one enterprise-grade at sub-800 ms latency, one on an open-source stack at a fraction of the cost.
  • ✓Week 6. A four-agent research swarm and "Jarvis", a desktop assistant for calendar, email and files on near-zero API spend.
  • ✓Week 7. Your digital twin plugged into a pipeline that scripts, renders and stages posts across platforms.
  • ✓Week 8. A capstone product, deployed and pitched live to guest judges with an architecture diagram and a business case.

How a weekend runs

Two sessions, three hours each.

01
The mechanism

How the thing actually works underneath — token sampling, distance metrics, agent loops, audio pipelines.

02
The architecture

How engineers assemble it in production: the trade-offs, the failure modes, the cost decisions.

03
The live build

We build it together, on screen, start to finish. When the hour ends, it runs on your machine.

Curriculum

8 weeks · 16 sessions · 15 builds + capstone
W01
LLM foundations and the four pillars of AI engineering
Model reasoning, tokenomics, and programmatic control of AI.
⌄
Saturday 3h live
Core mechanics and the four control pillars
  • Token processing, probability sampling, temperature, and why hallucinations happen
  • The four control pillars: prompt, context, harness and loop engineering
Build. A Claude Code skill that turns your weekly code activity into technical posts
Sunday 3h live
Token economics and constraint-based reasoning
  • Token economics, context-window budgeting, and enforcing structured JSON output
  • Constraint-based reasoning across dates, budgets, transit buffers and preferences
Build. The reasoning engine of an AI travel planner, turning plain English into validated itinerary data
W02
Enterprise RAG and knowledge graphs
Turning messy corporate PDFs into queryable intelligence.
⌄
Saturday 3h live
Vector search and enterprise RAG
  • Vector embeddings, distance metrics, and chunking strategies that survive real documents
  • Vector databases, metadata filtering, hybrid search and re-ranking pipelines
Build. An HR and finance assistant that answers from internal PDFs with page-level citations
Sunday 3h live
Knowledge graphs and GraphRAG
  • Where naive RAG breaks: global reasoning, relationship tracking, multi-hop queries
  • Entities, relationships and semantic triples; GraphRAG with Neo4j and NetworkX
Build. A knowledge graph mapping org hierarchies and financial interdependencies
W03
AI-assisted web coding and one-click deployment
Turning plain-English specs into deployed production software.
⌄
Saturday 3h live
The anatomy of modern software
  • Front ends, back ends, REST APIs, databases and servers
  • Git without terminal friction: staging, commits, branches, pull requests, GitHub
Build. Direct Claude Code or Cursor in natural language to scaffold and style your portfolio site
Sunday 3h live
Wiring UI to APIs and shipping it
  • Wiring UI to API endpoints, state management, and streaming tokens in real time
  • One-click hosting on Vercel or Render, custom domains, and safe environment variables
Build. Ship the Week 1 travel planner as a React front end on a live public URL
W04
Tool-calling agents and the Model Context Protocol
From passive chat interfaces to systems that execute.
⌄
Saturday 3h live
Agentic loops and MCP
  • Agentic decision loops: ReAct, tool schemas, argument validation, error recovery
  • MCP as a standard for connecting models to your OS, databases and dev environment
Build. Your own MCP server with authenticated read and write access to local files
Sunday 3h live
Web extraction and document agents
  • Programmatic web extraction and parsing job-board structures that keep changing
  • Generating and formatting PDF resumes against a specific job description
Build. An agent that pulls live listings, tailors a resume per company, and files each one in its own folder
W05
Real-time voice agents
Enterprise pipelines, and the same result on an open-source stack.
⌄
Saturday 3h live
The enterprise voice chain
  • Voice activity detection, speech-to-text, reasoning, speech synthesis
  • Full-duplex WebSockets, barge-in interruption handling, premium synthesis with ElevenLabs
Build. A conversational support agent responding in under 800 milliseconds
Sunday 3h live
Voice on an open-source budget
  • Where the cost actually goes, and driving it down to a rupee or two a minute
  • Local Whisper transcription, Cartesia and FastAPI streaming, lightweight audio pipelines
Build. A sub-second voice assistant for phone or browser with no SaaS lock-in
W06
Multi-agent orchestration and building "Jarvis"
Designing agent teams, and a desktop assistant that costs almost nothing to run.
⌄
Saturday 3h live
Agent teams that don't drift
  • Why single agents fail: context pollution, execution drift, hallucination loops
  • Supervisor-worker topologies, role routing and state-machine handoffs with LangGraph and CrewAI
Build. A four-agent research swarm: planner, researcher, fact checker, writer
Sunday 3h live
Building Jarvis
  • Combining MCP tools, shell execution and open-weight or free-tier model routing
  • Safe sandboxes, human-in-the-loop approval triggers, and background task management
Build. "Jarvis", a desktop assistant handling calendar bookings, email drafts and file operations
W07
Digital twins and the multi-channel brand engine
Scaling your authority through voice cloning and automated distribution.
⌄
Saturday 3h live
Voice cloning and synthetic presenters
  • Voice cloning mechanics: training data, audio normalisation, matching vocal inflection
  • AI video generation, lip-sync, and synthetic presenter workflows
Build. Your digital twin, delivering a technical explanation in your voice and likeness
Sunday 3h live
The multi-channel brand engine
  • Hook formulas, technical storytelling structures, and how distribution algorithms behave
  • Programmatic posting pipelines for LinkedIn, X and Instagram
Build. One topic in, a scripted video of your twin staged across every platform out
W08
Capstone hackathon and demo day
Turning eight weeks of skills into one pitchable product.
⌄
Saturday 3h live
Architecture review and build sprint
  • Architecture review: bottlenecks, database indexing, latency optimisation
  • Mentored build sprint with one-to-one office hours and live debugging
Build. Capstone architecture diagram and a deployed first version
Sunday 3h live
Demo day
  • Live pitches: each participant demos their deployed app, architecture and business case
  • Review by guest industry judges, prizes, and portfolio verification
Capstone. Pitch your capstone to guest judges

Who this is for

  • ✓ Engineers who can read code and want to move into AI work
  • ✓ Career switchers willing to spend both weekend days building
  • ✓ Product and data people who want to stop waiting on an engineer
  • ✓ Founders prototyping AI products without a technical co-founder

Who this is not for

  • ✕ Anyone looking for a passive video library to watch later
  • ✕ Anyone who wants theory without opening a terminal
  • ✕ Anyone who can't commit six hours across the weekend
  • ✕ Anyone expecting a job guarantee — you get proof of work instead

Who teaches it

K
Karthik
Lead instructor
Lead AI/ML Engineer (MLOps), UPS

Thirteen years across DevOps, MLOps, cloud infrastructure and agentic AI, with earlier roles at Amazon, Tabsquare (Delivery Hero) and Blackstraw.

P
Pradeep Boopathy
Instructor
Software and AI Engineer, Kaggle Master

Enterprise experience across UPS, CoreStack, TCS and Atos, and an active open-source contributor focused on making AI reliable and fast under production load.

V
Venkatesh Paramasivan
Instructor
AI/ML and Generative AI Specialist, Enterprise Systems Lead

Over twelve years in enterprise IT, postgraduate credentials in AI/ML, and more than ten end-to-end production deployments across retail and supply chain.