.png)
Forward Deployed AI Engineering for Banking
Certification Covered
Programme certificate · capstone panel defence
Stack
Program Includes:
Description
Forward Deployed AI Engineering for Banking is a 60-hour programme for software, platform, data and ML engineers who will design, build and operate agentic AI inside a bank. The ten-stage Agentic Delivery Lifecycle (ADLC) is the backbone: every session produces the artefact its stage demands, plus one Architecture Decision Record. Through 20 hands-on labs on synthetic banking data, you will build with Microsoft Agent Framework and Foundry, ground decisions with Azure AI Search, adapt models with Foundry fine-tuning, and govern against FCA, PRA, EU AI Act and DORA expectations. Every section starts with a foundational topic, and the programme ends with a capstone panel defence.
Course Outcomes
Upon completion, participants will be able to:
- Run decision discovery with bank stakeholders and turn an ambiguous ask into a Decision Card, decision model and ontology
- Design KPI trees and evaluation sets in Foundry before building, and gate every change on them in CI
- Build governed agents and multi-agent workflows with Microsoft Agent Framework, API Management tool contracts and middleware-enforced rules
- Ground decisions in policy with Azure AI Search and defend against hallucination and prompt injection
- Decide when fine-tuning is justified, then run and evaluate SFT, DPO or distillation in Foundry
- Deploy, observe and operate agents inside bank network controls, with runbooks and DORA-style incident handling
- Govern agents against FCA, PRA and EU AI Act expectations, and defend autonomy, controls and value before an executive panel
Prerequisites
- Programming: Working proficiency in Python. C# is accepted, as Agent Framework has .NET parity.
- APIs and tooling: Comfort with REST APIs, JSON, Git, containers and basic SQL.
- Azure basics: Resource groups, managed identities and role assignments.
- No prior LLM experience required: LLM and agent fundamentals are covered in Section 1.2, every section opens with a foundational topic, and a 25-question pre-assessment calibrates pace and pairing.
- Lab environment: A dedicated Azure sandbox per pair (Foundry project, Azure AI Search, Fabric workspace, API Management), provisioned before Session 1.
- Data rules: Only synthetic card, account, transaction and dispute data is used. Participants must not bring internal bank data into the environment.
Course Content
1The Forward Deployed Engineer in Banking & the Microsoft Agentic Landscape3 Sections
1.1 — The FDE Role and Agentic AI as Decision Automation
1.1 — The FDE Role and Agentic AI as Decision Automation
1.2 — LLM & Agent Essentials for Engineers
1.2 — LLM & Agent Essentials for Engineers
1.3 — The Microsoft Agentic Stack and Sandbox Orientation
1.3 — The Microsoft Agentic Stack and Sandbox Orientation
2Define & Model — Decision Discovery, Decision Cards & Decision Modelling2 Sections
2.1 — Decision Discovery and the Decision Card
2.1 — Decision Discovery and the Decision Card
2.2 — Decision Modelling and the Ontology
2.2 — Decision Modelling and the Ontology
3Measure & Context I — KPI Trees, Evaluation Sets & Agent-Ready Data2 Sections
3.1 — KPI Trees and Evaluation Sets in Foundry
3.1 — KPI Trees and Evaluation Sets in Foundry
3.2 — Agent-Ready Data on Microsoft Fabric
3.2 — Agent-Ready Data on Microsoft Fabric
4Context II & Build I — Tool Contracts, API Management & Agent Framework Fundamentals2 Sections
4.1 — Tool Contracts and Azure API Management as the Tool Gateway
4.1 — Tool Contracts and Azure API Management as the Tool Gateway
4.2 — Agent Framework Fundamentals and Model Selection
4.2 — Agent Framework Fundamentals and Model Selection
5Build II — Multi-Agent Workflows, Decision Records & AI-Native Engineering2 Sections
5.1 — Multi-Agent Workflows with Agent Framework
5.1 — Multi-Agent Workflows with Agent Framework
5.2 — Decision Records, AI-Native Engineering and the CI Evaluation Gate
5.2 — Decision Records, AI-Native Engineering and the CI Evaluation Gate
6Ground — Retrieval, Grounding Verification & Hallucination Control2 Sections
6.1 — Retrieval on Azure AI Search
6.1 — Retrieval on Azure AI Search
6.2 — Grounding Verification, Content Safety and Red-Teaming
6.2 — Grounding Verification, Content Safety and Red-Teaming
7Operate — Deployment, Observability & Evaluation in Production2 Sections
7.1 — Production Deployment and Observability
7.1 — Production Deployment and Observability
7.2 — Operating Modes, Incident Response and Resilience
7.2 — Operating Modes, Incident Response and Resilience
8Feed Back & Govern — Closing the Loop, Adapting Models & Governing Decisions3 Sections
8.1 — Feedback Loops from Human Overrides
8.1 — Feedback Loops from Human Overrides
8.2 — Governance, Agent Identity and Consumer Duty
8.2 — Governance, Agent Identity and Consumer Duty
8.3 — Model Adaptation: When and How to Fine-Tune
8.3 — Model Adaptation: When and How to Fine-Tune
9The Copilot Layer — Copilot Studio, Microsoft 365 Copilot & Pro-Code Integration2 Sections
9.1 — Copilot Studio and Microsoft 365 Copilot Extensibility
9.1 — Copilot Studio and Microsoft 365 Copilot Extensibility
9.2 — Connecting Pro-Code Agents and Choosing the Surface
9.2 — Connecting Pro-Code Agents and Choosing the Surface
10Prove — Capstone Build & Panel Defence2 Sections
10.1 — Capstone Build and Value Proof
10.1 — Capstone Build and Value Proof
10.2 — Panel Defence and Peer Review
10.2 — Panel Defence and Peer Review
Overview of Capstone Projects
Your Banking Decision, End to End on Microsoft Foundry
Your Banking Decision, End to End on Microsoft Foundry
- Objective: Working in pairs, participants apply the full ADLC to a banking decision of their own choosing — never the card dispute running example — then build, deploy, evaluate and defend it before a panel playing Chief Risk Officer, Head of Operations and Head of Engineering.
- Capstone menu (chosen in Session 8):
- KYC document review — multimodal document extraction with abstain and escalate paths.
- AML alert triage — evidence gathering and disposition recommendations with human sign-off.
- Collections next-best-action — Consumer Duty vulnerability routing and fair outcomes.
- Credit limit decisioning — an EU AI Act high-risk use case with GDPR Art. 22 explanation duties.
- Complaints handling under Consumer Duty — root-cause analysis and response drafting.
- Build requirements:
- Define and model: Decision Card, DMN-style decision tables and ontology.
- Measure and context: KPI tree, labelled evaluation set registered in Foundry, and a Fabric data product with Purview labels.
- Build: tools published through API Management, an Agent Framework workflow with middleware-enforced rules, and a CI evaluation gate.
- Ground and adapt: Azure AI Search retrieval with citations, groundedness checks, and a documented decision on whether fine-tuning is justified.
- Operate and govern: deployment with OpenTelemetry tracing, decision log, runbook and a governance register.
- Deliverables:
- Deployed system — running on Foundry, with its evaluation report by dimension
- Value proof — baseline vs agent on the KPI tree: cost per decision, time to decision, override and escalation rates
- Artefact set — Decision Card, decision model, ontology, KPI tree, evaluation set, tool contracts, workflow, deployment and governance register
- ADR decision log — the ten Architecture Decision Records written across the programme
- Panel defence — a 15-minute defence scored by panel and peers, with individual written feedback
- Assessment and certification:
- Pre-assessment — 25 scenario questions on agentic concepts, decision modelling, Azure and LLM basics.
- In-class checks — scenario checks at the close of each ADLC stage, plus one coach-reviewed ADR per session.
- Post-assessment — scenario assessment across all ten ADLC stages plus the capstone.
- Certification criteria — 90% attendance, all labs completed, full ADR set submitted, capstone deployed with evaluation results and defended before the panel.
Tech Stack
Career Opportunities
Learning Roadmap
The Forward Deployed Engineer in Banking & the Microsoft Agentic Landscape
3 Sections
Define & Model — Decision Discovery, Decision Cards & Decision Modelling
2 Sections
Measure & Context I — KPI Trees, Evaluation Sets & Agent-Ready Data
2 Sections
Context II & Build I — Tool Contracts, API Management & Agent Framework Fundamentals
2 Sections
Build II — Multi-Agent Workflows, Decision Records & AI-Native Engineering
2 Sections
Ground — Retrieval, Grounding Verification & Hallucination Control
2 Sections
Operate — Deployment, Observability & Evaluation in Production
2 Sections
Feed Back & Govern — Closing the Loop, Adapting Models & Governing Decisions
3 Sections
The Copilot Layer — Copilot Studio, Microsoft 365 Copilot & Pro-Code Integration
2 Sections
Prove — Capstone Build & Panel Defence
2 Sections
Certified Forward Deployed AI Engineering for Banking
Goal Achieved
Weekdays (Mon-Fri)
Starting from 16th November
Duration: 2 Hours per day
Weekend (Sat & Sun)
Starting from 21st November
Duration: 4 Hours per day
Request For Demo
How do I enroll in this course?
Explore
Explore the course and certification
Consult
Talk to our agents for guidance
Experience
Join our live demo session
Enroll
Complete your enrollment
Meet Our Leaders
Program Director
Set the direction of the program and ensure it meets industry and academic excellence standards.
Aashish Pandey
Blue Data Consulting
Program Manager
Manages schedules, coordination, assessments, and ensures smooth course delivery.
Ragini Sankhwar
Blue Data Consulting
Industry Expert/Trainers
Harsh Dalal
Blue Data Consulting
AI Consultant
SME
Covers theory, hands-on labs, assessments, and daily assignments.
Aproova
Blue Data Consulting
Master SME
Specialized guidance for Capstone Projects and advanced implementation.
Ahmed
Blue Data Consulting
Mentor
Dedicated support to help all learners pass the certification exam.
Transparent Pricing
Invest in your future with our flexible payment options
Application Fee
5% of Total Fee (Non-refundable)
Program Fee
Complete program access
(After Application Fee + 10% Discount)
*18% GST & Certification cost included
*Important: First pay Application Fee, then you can enroll in Program
Easy EMI Plans
Pay in flexible monthly installments
Starting from
Frequently Asked Questions
Who is the target audience for this course?
The course is for software, platform, data, and ML engineers who will design, build, and operate agentic AI inside a bank.
What technologies are covered?
The stack includes Microsoft Agent Framework, Microsoft Foundry, Azure OpenAI, Azure AI Search, Microsoft Fabric, API Management, Entra Agent ID, and Copilot Studio.
How long is the programme?
It is a 60-hour programme consisting of 10 sessions that are about 6 hours each.
Do I need prior LLM experience to enroll?
No prior LLM experience is required. However, working proficiency in Python (or C#) and comfort with REST APIs, JSON, Git, and Azure basics are required.
Are there practical exercises?
Yes, the course includes 20 hands-on labs. Participants also receive one dedicated Azure sandbox per pair.
Will we use real banking data during the labs?
No, only synthetic card, account, transaction, and dispute data is used. Participants must not bring internal bank data into the environment.
What is the ADLC?
The Agentic Delivery Lifecycle (ADLC) is a ten-stage framework (Define, Model, Measure, Context, Build, Ground, Operate, Feed Back, Govern, Prove) that acts as the backbone of the course.
How is the course assessed for certification?
Certification requires 70% attendance, completion of all labs and Architecture Decision Records (ADRs), passing a post-assessment, and successfully deploying and defending a capstone project.
What does the capstone project entail?
Working in pairs, participants apply the full ADLC to a chosen banking decision (e.g., KYC review, AML triage, credit limit decisioning) and defend their deployed system before a panel playing the Chief Risk Officer, Head of Operations, and Head of Engineering.
How does the course handle AI regulations?
Regulatory expectations from the FCA, PRA, EU AI Act, and DORA are taught as engineering design constraints to help govern the agents.