✨Updates to Bootcamp!
Curriculum

Learn by Building

Build production-ready AI systems first. Learn the theory as you go.

Course Outline (Project-Based Learning)

1

Build Your Own Perplexity

Create a production-ready Retrieval-Augmented Generation (RAG) system with document uploads, web search, evaluations, and observability.

RAG Pipeline Foundations

  • •Implement document upload, parsing, and indexing
  • •Create vector embeddings and store in a vector database
  • •Retrieve relevant context with source citations

Dynamic Web RAG

  • •Integrate ethical web search and domain-specific crawling
  • •Add caching and deduplication for better performance
  • •Optimize retrieval speed and freshness of search results

RAG Evaluation & Observability

  • •Run RAG metrics like faithfulness, precision, and recall
  • •Build LLM-as-a-judge test suites for regression testing
  • •Add full tracing, logging, and latency monitoring
Image Placeholder
Add your project image here
2

Multi-Agent Content Automation

Build a multi-agent system using MCP servers and A2A protocol to research, write, and publish automated content.

Agent Architecture

  • •Design planner, researcher, writer, reviewer, and publisher agents
  • •Use MCP servers to access third-party tools and APIs
  • •Implement A2A protocol for agent-to-agent collaboration

Agent Reliability

  • •Add retries, circuit breakers, and tool-call fallbacks
  • •Track tool-call accuracy, failures, and latency
  • •Build observability dashboards for agent traces

Agent Evaluation

  • •Measure agent correctness and task completion rate
  • •Add content safety evaluations and compliance rules
  • •Integrate agent-level CI gates for reliability
Image Placeholder
Add your project image here
3

Multimodal Asset Factory

Build multimodal agents that generate high-quality images, videos, and voice assets for marketing and product storytelling.

Image & Video Generation

  • •Generate hero graphics and short promotional videos
  • •Apply style prompts for brand consistency
  • •Filter and validate safe content during generation

Audio & Voice Generation

  • •Generate voiceovers and narration using TTS models
  • •Ensure clarity and tone consistency
  • •Use ASR-based evaluation to detect quality issues

Multimodal Evaluations

  • •Run CLIPScore, FID/FVD, and VBench metrics
  • •Use LLM-as-a-judge for creative evaluations
  • •Add regression tests to prevent quality degradation
Image Placeholder
Add your project image here
4

Host Your Own Small Language Model

Deploy and fine-tune a small language model using vLLM, LoRA/QLoRA adapters, and GPU-based inference optimization.

vLLM Model Serving

  • •Deploy vLLM on GPU servers or Kubernetes
  • •Serve multiple LoRA adapters dynamically
  • •Optimize inference speed and GPU memory usage

Model Fine-Tuning

  • •Generate domain-specific synthetic training data
  • •Train LoRA/QLoRA adapters for targeted tasks
  • •Evaluate fine-tuned model quality vs. baseline

Cost Optimization & Routing

  • •Route simple queries to your SLM for savings
  • •Measure cost-per-query and latency improvements
  • •Monitor GPU utilization and model performance
Image Placeholder
Add your project image here
5

Self-Improving Agents

Build agents that learn from feedback using reinforcement learning, bandit optimization, and continuous evaluation.

Feedback Loop Design

  • •Collect and process user ratings and signals
  • •Design safe reward functions for improvement
  • •Integrate online learning based on real usage

Reinforcement Learning Basics

  • •Use contextual bandits for prompt optimization
  • •Run offline RL simulations using historical data
  • •Apply conservative updates to avoid regressions

Continuous Evaluation

  • •Monitor regressions using automated evals
  • •Perform canary and shadow testing before deployment
  • •Ensure safety, reliability, and latency remain stable
Image Placeholder
Add your project image here
6

Your Own Production AI Application

Apply everything you've learned to build your own production-ready AI system. Choose a real-world problem, design the architecture, and ship a complete solution using techniques from all previous projects.

Project Guidelines

  • •Choose your own use case and problem to solve
  • •Define scope and technical requirements
  • •Get instructor feedback and guidance throughout development

Technical Requirements

  • •Apply techniques from Projects 1-5 as needed
  • •Production-ready architecture with proper error handling
  • •Full observability, monitoring, and logging implementation
  • •Cloud deployment with Infrastructure-as-Code (Terraform)
  • •CI/CD pipeline with automated testing

Deliverables & Presentation

  • •Working application deployed to your chosen cloud platform
  • •Complete documentation and technical specification
  • •Live demo presentation to cohort and instructors
  • •Code review session with detailed architectural walkthrough
Image Placeholder
Add your project image here

Cloud Platform Support

All projects are cloud-agnostic and can be deployed on AWS, Azure, GCP, Databricks, or Snowflake with included Infrastructure-as-Code (Terraform) and CI/CD pipelines.