Build and Deploy Agentic AI Apps with CrewAI, Autogen, ADK and Streamlit

Official course repository for the WSQ course Build and Deploy Agentic AI Apps with CrewAI, Autogen, ADK and Streamlit (TGS-2025059028), delivered by Tertiary Infotech Academy Pte Ltd (UEN 201200696W).

Live demo: Invoice RAG Chat

---

What you build

Twenty hands-on activities take you from a single tool-using agent to a publicly deployed, guardrailed multi-agent application.

#	Application	Framework
3	Chat agent with live web search	OpenAI Agents SDK + Tavily
4	Role-based content crew (Researcher → Writer)	CrewAI
5	Conversational writer/critic team (reflection)	Microsoft AutoGen
6	Multi-agent trip planner (agents-as-tools)	OpenAI Agents SDK
8	Invoice RAG over PDFs	ChromaDB + embeddings
12	Public cloud deployment	Streamlit Community Cloud
16–17	Input + output guardrails with tripwires	OpenAI Agents SDK guardrails
20	Guardrailed production app	everything above

Course structure

Learning Unit	Focus	Activities
LU1 — Agentic AI App Development	Frameworks, multi-agent workflows, workflow design, AI IDEs	1–7
LU2 — RAG & Deployment on Streamlit Cloud	Context augmentation, knowledge graphs, cloud inference, error logs	8–14
LU3 — Model Alignment and Guardrails	Safeguarding LLMs, alignment, deploying guardrails	15–20

Learning outcomes

1. Design and implement multi-agent integration workflows using CrewAI, AutoGen and the OpenAI Agents SDK / ADK.
2. Develop and deploy Agentic AI and RAG applications for inference on Streamlit Cloud.
3. Deploy guardrails and model alignment techniques to ensure safe and aligned outputs.

Aligned to Skills Framework TSC Generative AI Application Development and Deployment (ICT-INT-0047-1.1).

---

Architecture

                        ┌──────────────────────────┐
   user request ───────▶│    INPUT GUARDRAIL       │  classifier agent + tripwire
                        │  injection · off-topic   │
                        └────────────┬─────────────┘
                                     │ passes
                        ┌────────────▼─────────────┐
                        │       ORCHESTRATOR       │
                        │   (agents-as-tools)      │
                        └───┬─────────┬─────────┬──┘
                   ┌────────▼──┐ ┌────▼─────┐ ┌─▼─────────┐
                   │ Planner   │ │ Budget   │ │ RAG /     │
                   │ agent     │ │ agent    │ │ retrieval │
                   └────┬──────┘ └────┬─────┘ └─────┬─────┘
                        │  tools      │             │  ChromaDB
                        └─────────────┴─────────────┘
                                     │
                        ┌────────────▼─────────────┐
                        │    OUTPUT GUARDRAIL      │  PII redaction + policy check
                        └────────────┬─────────────┘
                                     ▼
                              answer to user
                        (every decision → audit log)

Repository layout

labs/
  lab-01-set-up-the-agentic-ai-development-environment/
    README.md          the goal, the workflow, every step, the "Test it" check
    app.py             the runnable app  (streamlit run app.py | python app.py)
    requirements.txt   only what THIS lab needs
    data/              any CSV or sample PDFs the lab uses
  lab-02-compare-crewai-autogen-and-the-openai-agents-sdk/
  lab-03-build-a-single-agent-with-tools-openai-agents-sdk/
  …                    20 self-contained lab folders in all
  README.md            the lab index, grouped by Learning Unit
  .env.example         copy to .env at the labs root and add your own keys
courseware/            slide deck (202 slides), Lesson Plan, Learner Guide (+ PDFs)

Every lab folder is self-contained: read its README, install its own requirements.txt, run its app.py. One shared .env at the labs root supplies the API keys to all of them.

Setup

git clone https://github.com/tertiarycourses/TGS-2025059028-Build-and-Deploy-Agentic-AI-Apps-with-CrewAI-Autogen-ADK-and-Streamlit.git
cd TGS-2025059028-Build-and-Deploy-Agentic-AI-Apps-with-CrewAI-Autogen-ADK-and-Streamlit

python3 -m venv .venv && source .venv/bin/activate     # Windows: .venv\Scripts\activate

Then install per lab — each folder declares only what it needs:

pip install -r labs/lab-03-build-a-single-agent-with-tools-openai-agents-sdk/requirements.txt

Create your .env from the template and paste in your own keys:

cp labs/.env.example labs/.env

Key	Where to get it	Free tier
OPENAI_API_KEY	platform.openai.com	pay-as-you-go
TAVILY_API_KEY	app.tavily.com	yes
GEMINI_API_KEY	aistudio.google.com	yes
OPENROUTER_API_KEY	openrouter.ai	yes

Never commit .env. It is git-ignored. For Streamlit Cloud, paste the same keys into the app's Secrets manager instead (see labs/.streamlit/secrets.toml.example).

The current trainer deck is courseware/Build and Deploy Agentic AI Apps with CrewAI, Autogen, ADK and Streamlit-v2.pptx (202 slides).

Run

Each lab runs from its own folder:

cd labs/lab-03-build-a-single-agent-with-tools-openai-agents-sdk
pip install -r requirements.txt
streamlit run app.py                         # Activity 3 — chat agent with web search

cd labs/lab-04-build-a-role-based-crew-with-crewai   && python app.py
cd labs/lab-05-build-a-conversational-team-with-autogen && python app.py
cd labs/lab-06-orchestrate-a-multi-agent-trip-planner-agents-as-tools && streamlit run app.py
cd labs/lab-08-build-a-rag-pipeline-over-invoice-pdfs && streamlit run app.py
cd labs/lab-16-implement-input-guardrails-with-tripwires && streamlit run app.py

Deploy to Streamlit Community Cloud

1. Confirm requirements.txt lists every import — a missing line is the most common failure.
2. Verify no key is committed: git grep -nE '(sk-|tvly-)[A-Za-z0-9]{20}'
3. Push to a public GitHub repo.
4. At share.streamlit.io → Create app → select the repo, branch and main file.
5. Advanced settings → Secrets, paste your keys in TOML format.
6. Deploy, then open the public *.streamlit.app URL.

Courseware

Slides, the Lesson Plan and the Learner Guide are distributed to enrolled learners through lms-tms.tertiaryinfotech.com. Assessment material is confidential and is not published in this repository.

Acknowledgements

Built with CrewAI, Microsoft AutoGen, the OpenAI Agents SDK, Streamlit, ChromaDB and Tavily.

License

© 2026 Tertiary Infotech Academy Pte Ltd (UEN 201200696W). All rights reserved. Provided for the use of enrolled learners of TGS-2025059028.