Mastering Agentic AI
An 8‑week, hands‑on live cohort. Move from “just prompting” to
designing and building real agent workflows end‑to‑end.
- Design multi-agent workflows: connect multiple AI agents that collaborate on realistic enterprise tasks
- Integrate tools and data (RAG + APIs): safely plug AI into internal docs, knowledge bases, and selected APIs
- Add guardrails and evaluation: test, debug, and monitor agent behaviour so it’s safer and more reliable before you show it to others
Starts 12th September 2026.
Start date: 12 September 2026 (live weekends)
Capstone project: Build “AskFriday”
A practical multi-agent assistant built step-by-step across 8 weeks that you can demo to your team or in interviews.
Our Differentiation
Eight weeks of live mentorship, hands-on building, and shipping a production-ready multi-agent system.
Builders, Not Generic Trainers
Live sessions taught by engineers and scientists who are building AI systems today at companies like Microsoft, HPE, Adobe, etc. not full-time “course instructors.”
Public Portfolio
We will help you build or improve your public portfolio on LinkedIn/GitHub, including your unique Pitch pack for the multi-agent you build.
Career support
Access to a closed community of peers and mentors for 3 months after the program, including monthly office hours and a space to discuss real‑world challenges.
8-Week Build Plan
Every week is structured around practical deliverables. You'll progress from basics to deploying sophisticated multi-agent structures.
Async Foundations & Your First Agent
- Configure a modern Python dev environment for scalable, non-blocking AI workflows.
- Learn async Python patterns for running multiple LLM calls and tools efficiently.
- Build a tool-calling agent using Groq + Tavily web search with Pydantic-structured outputs.
- Deploy your first agentic app on HuggingFace Spaces.
Project: AskFriday v0: a research agent deployed on HuggingFace Spaces.
RAG - Vector Search, Corpus Engineering & Ragas Metrics
- Implementing fixed-size, semantic, and recursive chunking strategies over a synthetic Indian enterprise corpus
- Building dense, sparse, and hybrid retrieval pipelines
- Storing and querying embeddings in ChromaDB, with an optional Pinecone upgrade path
- Refining retrieved results through query rewriting, reranking, and contextual compression (Context Engineering)
- Building an automated Ragas evaluation pipeline to track retrieval faithfulness and answer accuracy
- NVIDIA Resources: NVIDIA NV-Embed (via NIM API), NVIDIA NeMo Reranker
Project: AskFriday v1 - Document Q&A agent that retrieves answers from an enterprise document corpus using ChromaDB, with retrieval quality continuously measured through Ragas metrics.
LangChain Fundamentals
- Composing prompts, models, and parsers into declarative pipelines using LCEL (LangChain Expression Language)
- Designing reusable Prompt Templates to standardize and parameterize model instructions
- Enforcing structured, validated model outputs using LangChain's Output Parsers, including JSON schema and Pydantic-based parsing
- Building Chains and Runnables to compose multi-step LLM workflows in a consistent, async-compatible interface
- Wrapping retrieval pipelines and external tools as LangChain Retrievers and Tool Calling interfaces
- Managing conversational and workflow state using LangChain's Memory abstractions
- Understanding why LangChain and LangGraph exist as complementary layers, and when to reach for each
Project: AskFriday v2 - Week 2's RAG pipeline re-expressed using LangChain: prompts, parsers, and retrieval composed as LCEL chains and Runnables.
LangGraph - Stateful Agents with Memory
- Building a LangGraph state machine with a retrieve-vs-act router and conditional routing logic
- Adding a reflection node so the agent can critique and revise its own output
- Using LangGraph's time-travel debugging to inspect and replay past agent states
- Implementing a Postgres-backed checkpointer, hosted on a managed service (Supabase), for durable cross-session memory
- Adding human-in-the-loop approval steps using LangGraph interrupts, pausing execution for human review before continuing
- NVIDIA resources: LangGraph optimization patterns, Nemotron via NIM API.
Project: AskFriday v3 - Stateful router agent that persists memory across sessions using a cloud-hosted Postgres checkpointer, with human-in-the-loop approval points and no local database setup required.
MCP Protocol Design & Custom Tool Server Engineering
- Learning the Model Context Protocol (MCP) as a standard way for LLMs to securely call external tools
- Building a custom MCP server using the MCP Python SDK and FastAPI
- Exposing a mock ERP system as a set of MCP tools that any compliant agent can call
- Publishing and versioning the MCP server so it can be reused across agents
- Choosing between reasoning and fast models for tool-selection decisions (model selection strategy)
- Handling failed or malformed tool calls with retry strategies and production failure handling
- NVIDIA resources: Nemotron Mini 4B via NIM API.
Project: AskFriday v4 - Tool-using agent that calls a custom MCP server to interact with a mock ERP system, with retry logic for failed tool calls.
Multi-Agent Systems & Production Telemetry
- Designing a four-agent LangGraph graph combining “Orchestrator-Worker” and “Planner-Actor-Critic” patterns
- Instrumenting the multi-agent system with live Langfuse tracing for full step-by-step observability
- Benchmarking the system against a 50-question golden dataset using LLM-as-judge evaluation
- Adding NeMo Guardrails for PII redaction and topic control across all agents
- Comparing LangSmith and Langfuse for agent observability and evaluation
- Running regression tests and structured prompt evaluations as the system evolves
- Weighing cost vs. latency trade-offs across model and API choices
- Debugging failed tool calls and agent loops in a multi-agent setting
- NVIDIA resources: NeMo Data Designer, NeMo Guardrails.
Project: AskFriday v5 - Four-agent system with Indian-language support via Sarvam-M, fully traced in Langfuse, regression-tested, and evaluated against a golden dataset.
Ship, Govern & Present - From Docker to Demo
- Packaging the full multi-agent stack into reproducible Docker containers for production deployment
- Implementing cost governance controls across all NIM and API-based model calls
- Writing Architecture Decision Records (ADRs) comparing NIM Cloud, cloud GPU, and on-premises infrastructure paths
- Preparing and delivering a Demo Day presentation to an industry jury
Project: AskFriday v6 - Fully containerized capstone system, presented live to an industry jury together with supporting documentation and ADRs.
Bonus session - Career Accelerator
Bonuses
Extras that help you turn what you build into career leverage.
Portfolio & Internal Pitch Pack
LinkedIn Playbook (8 weeks)
NVIDIA-Enabled Starter Stack
6-Month Access to Recordings & Materials
Live Expert Sessions
Every week, bring your questions to live sessions with experts from companies like HPE, Microsoft, Adobe, or Accenture. Get help with implementation, discuss your Project patterns, or gain direct insights on career growth.
HPE
Microsoft
Adobe
AccentureFrequently Asked Questions
Basic Python proficiency is all you need. For those who need help with, we share short prep guides & LIVE prep session before Week 1.
Still have questions? Drop us a note at hello@thefriday.ai
Mastering AI Agents
September 2026 Batch Enrollment
Ready to build production-grade AI agents?
September 2026 cohort