AI Learning Assistant for a Global Bank (NDA)

We helped a global bank’s corporate education team create a secure AI assistant that makes learning faster and everyday work easier.

The solution brings together department-specific knowledge, a multi-LLM chatbot, voice and image features, and program-level study assistants powered by RAG and all deployed inside the bank’s closed infrastructure.

The challenge

The bank runs many training programs for different departments with strict security rules. Content is large, changes often, and employees ask similar questions again and again.

The client needed one simple entry point where staff could:

get trustworthy answers from internal materials,

practice using modern AI safely,

and speed up course creation without breaking any policies.

Prompt practice (trainer)

A guided space to practice prompts safely. Employees learn how to ask better questions and get feedback, which improves real-world results.

Voice & image features

Speech-to-text and text-to-speech make the assistant accessible and fast to use during busy days by adding a voice interface. Image generation helps create simple visuals and examples for training modules.

Study program bots

For each training program, a small “study bot” analyzes lectures, handouts, and quizzes. Learners can ask questions about the course and get clear answers with references. This reduces waiting time for human instructors.

Secure multi-model chatbot

A familiar “ChatGPT-style” interface that runs fully inside the bank’s environment. It can route prompts to different local models depending on the task, while keeping data inside the perimeter.

Department assistants (RAG)

Each department (e.g., Compliance, Operations, Retail) has its own knowledge store with access controls. The assistant searches and cites internal documents, policies, slide decks, and FAQs, so answers are consistent and traceable.

What we built

Security 
and compliance

Deployed fully inside the bank’s infrastructure under stringent cybersecurity and architecture standards.

We worked with IT and Security through several internal reviews and audits, aligning to policies on access, data handling, change control, and monitoring.

We delivered the documentation and remediation needed to pass all gates for go-live.

Integrations

The assistant connects to the learning portal, document repositories, and internal authentication. Admins can manage content sources, measure usage, and view simple quality dashboards (answer coverage, top sources, feedback).

Rollout

We started with a short discovery and a focused pilot, then expanded department by department. We trained “champions” inside the client teams and collected feedback to improve prompts, sources, and UI. The system now serves thousands of employees across multiple regions.

Results

The assistant gave employees one secure place to get reliable answers and get work done faster. Training teams update and reuse materials more easily, and staff can practice modern AI safely under bank policies.

On-prem deployment and internal reviews satisfied security and architecture requirements. As it rolled out by department, usage grew to thousands of employees with clear reductions in repeat questions and turnaround time.

FastAPI

Kubernetes

Docker

Minio

React

Postgres

Technology Stack:

OpenAI API, vLLM

PGVector

LLM Tools

RAG Engine

STT, TTS, Image generation models