Case Study: Banking Multimodal RAG
A ~36-minute walk through a real bank multimodal RAG build — Azure AI Search + Vision + GPT-4o, document cracking, MRM submission. Voice: Andrew.
8
Chapters
~36 min
Duration
Advanced
Level
No
Certification
Who this is for
For technical leaders, ML engineers, and product managers learning by walking through a real bank multimodal RAG build.
How this course works
- 8 audio-narrated slide chapters · ~36 min of focused content
- Narrated by Andrew (Azure neural voice)
- Capstone with interactive Markdown builder you take to your team
- Trust trip-wires on every play — what not to cross
- Free verifiable certificate on completion
What you'll walk out with
Specific outcomes from this course — no fluff.
- Why pure text-RAG fails on financial documents — three concrete failure modes (tables, charts, scanned annexures) and four pre-committed success criteria
- A Stack Fit Assessment for Azure AI Search + Vision + GPT-4o mapped across a three-stage pipeline
- Document cracking discipline for text, tables (small whole / medium row-group / large individual-row), and charts (~25%→80% lift on chart queries)
- Nine-field index schema with embedding choice (text-embedding-3-large) and chunk-size discipline (banking-dense 400-800 tokens)
- Hybrid search tuned for financial vocabulary — 3-component (vector + BM25 + semantic ranker), top 30 → ranker → top 8 to LLM, ~600ms budget
- Three-layer evaluation (retrieval + grounding + answer accuracy) with 200-query gold set and CI hard-block on regression
- Five-section MRM submission + five regulator questions (RBI + SAMA + EU AI Act) + the parallel 6-week regulatory assessment
- Five lessons from the build + four trust trip-wires + interactive build-plan builder
Course content
8 chapters · ~36 min
Welcome
A 1-minute orientation — what the course covers, how to navigate, and what you walk out with. No audio on this screen.
The banking use case
3 text-RAG failure modes · 4 success criteria · the pivot to multimodal.
Stack Fit Assessment
3-stage pipeline (ingest+crack · index+retrieve · generate+cite) · Azure stack per stage.
Document cracking
3 content types · 3 crackers · table-cracking discipline · chart-cracking 25%→80% lift.
Indexing strategy
9-field schema · embedding choices (text-embedding-3-large) · chunk size discipline (400-800 tokens).
Hybrid search at production scale
3-component hybrid (vector + BM25 + semantic ranker) · financial-vocab tuning · top 30 → ranker → top 8.
Evaluation harness
3-layer eval (retrieval · grounding · answer accuracy) · 200-query gold set · CI integration with hard-block.
Regulatory acceptance
5-section MRM submission · 5 regulator questions · RBI + SAMA + EU AI Act specifics.
Capstone — Lessons learned
5 lessons from this bank's build · 4 trust trip-wires · interactive build-plan builder.
Want this delivered inside your organisation?
The course is the starting point. The same content powers a 4-week pilot, an org-wide rollout, or a continuous build engagement — set up on your data, with your team, by Gennoor Tech.