FNB-EAssist: Teller at your fingertips

Voice-Enabled Banking – iPad Concept: This mockup highlights a potential tablet layout with voice access. Final UI may vary
E-Assist - Mobile Concept: Mobile interface showcasing one of our possible design paths. Visual only and not reflective of final design.
FNB E-Assist – Desktop Concept: A potential layout for desktop use. This is a visual prototype and may differ from the final app.

Inspiration

Many FNB app users, especially the elderly, disabled, or those not tech-savvy, struggle to navigate the app and find what they need quickly. This leads to frustration and negative feedback. We wanted to create a solution that makes banking easier for everyone by adding an AI Virtual Teller that guides users, answers their questions, and offers helpful financial advice in a simple, friendly way.

What it does

Our AI Virtual Teller (e-Assist) allows users to ask questions about the FNB app or banking features through a text input. The system sends these questions to ChatGPT, which interprets what the user wants. Based on the AI’s structured response, the app directs the user to specific feature pages or provides general financial tips. This makes navigating the app more intuitive and accessible.

How we built it

We built a Python app with a basic graphical interface, including:

A Login page for user access.
A Main Menu with navigation buttons.
A Text Input screen to ask the AI teller.
Integration with the OpenAI ChatGPT API to process questions.
A response parser that reads ChatGPT’s reply to decide which page to show.
Five key feature pages (A through E) representing common user requests.
Optional chat history and feedback functions.

Challenges we ran into

Crafting prompts that reliably produce structured AI responses was difficult.
Parsing AI answers to correctly link to app pages required careful string handling.
None of us had built a full app or used AI APIs before, so we had a steep learning curve.
Making financial advice clear, friendly, and legally safe took several iterations.

Accomplishments that we're proud of

Successfully integrated AI chatbot technology into a functional app prototype.
Designed a smooth flow from user input through AI processing to relevant page display.
Created a system that makes banking easier and more accessible to diverse users.
Developed a solid base that can be expanded with more features and accessibility improvements.

What we learned

How to work with AI language models and use APIs effectively.
How to design user-friendly interfaces with clear logic flows.
The importance of prompt engineering when working with AI.
Collaborating under time constraints and rapidly acquiring new tech skills.

What's next for FNB-eAssist: Teller at your fingertips

Adding voice input and multilingual support for wider accessibility.
Expanding the number of features and refining AI prompts for better accuracy.
Creating a mobile app version with animations and enhanced user experience.
Integrating real-time account data securely for personalized assistance.
Continuing to gather user feedback and improve the system iteratively.

Built With

openai
python

Submitted to

FNB Botswana: Bank of the Future

Created by

**FNB Bank Hackathon Project Documentation**

**Project Title:** Intelligent AI Assistant for FNB Bank

**Team Role Description:** Programmer (Word Search & Recommendation Logic)

**Overview:**
Our team is participating in the FNB Bank Hackathon with an ambitious project to develop an intelligent AI assistant aimed at enhancing user interaction within the bank's digital ecosystem. The AI assistant is intended to guide users efficiently through FNB's services, helping them find the right tools and resources without navigating multiple interfaces manually.

As the programmer on the team, my primary responsibility is to architect and implement the core logic that powers the assistant's recommendation engine. My focus is on developing the search and matching algorithms that allow the assistant to interpret user queries and recommend the most appropriate functionality within the banking app.

**Core Concept:**
At the heart of the AI assistant lies a robust text processing engine that can understand user intent through keyword analysis and semantic matching. We are using classical Information Retrieval (IR) techniques, particularly the BM25 algorithm, which is an extension of the probabilistic ranking framework used to assess the relevance of documents (or, in our case, application modules) to a given query.

**How the Code Will Work:**

1. **Input Capture:**

* The AI assistant listens or reads user input, which may be a typed query or a voice-to-text transcription.
* Example input: "I want to open a new savings account"

2. **Text Preprocessing:**

* Tokenization: Splitting user input into meaningful keywords.
* Stopword Removal: Removing common words (e.g., "I", "a", "to") that do not contribute to semantic meaning.
* Normalization: Lowercasing, stemming, or lemmatizing words to a common form.

3. **BM25 Scoring Engine:**

* Each module or feature in the FNB app is indexed as a "document" with metadata and descriptions.
* The BM25 algorithm scores each document based on its relevance to the user query.
* Example: "Open a new savings account" would match highly with a module document labeled "Account Opening > Savings Account".

4. **Recommendation Output:**

* The top-scoring documents are presented as recommendations to the user.
* These may be displayed as clickable options or read aloud, depending on the interface.
* Additional logic may prioritize modules based on usage frequency, time of day, or personalized preferences.

5. **Feedback Loop:**

* User interaction with the recommended options can be logged to improve future predictions.
* A lightweight learning mechanism can be added to adjust weights or incorporate synonyms over time.

**Why BM25:**
BM25 is especially effective in ranking short user queries against a corpus of semi-structured documents. It provides:

* A probabilistic relevance model
* Tunable parameters (e.g., k1 and b) for controlling term frequency saturation and document length normalization
* Strong performance in sparse data situations, where user input is minimal

**Future Integration Plans:**

* Use of Named Entity Recognition (NER) to better understand financial terminology
* Integration with a voice assistant framework for hands-free banking
* Incorporation of user profiling for more personalized experiences

**My Contributions:**

* Designed and implemented the BM25-based search and ranking system
* Built the document corpus from structured app module metadata
* Developed preprocessing pipeline for text normalization and analysis
* Collaborated with the front-end and UI team to ensure seamless integration of recommendations
* Optimized scoring logic for speed and relevance

**Conclusion:**
Our AI assistant aims to revolutionize how users interact with banking services by making navigation intuitive and context-aware. My role as the programmer focused on search and recommendation ensures that the assistant can accurately understand and respond to user needs in real time, using proven techniques from information retrieval. This not only enhances the user experience but also showcases how classical algorithms like BM25 can play a powerful role in modern AI-driven applications.

---

*Prepared by: Shaduff Nkabule*
*Role: Programmer / AI Logic Developer*
*Team: FNB Bank Hackathon 2025*

Chebshe Nkabule
Adelle Maphosa
Immaculate .iphizothile sithole
altonmabina
Sheenah Nsubuga

Updates

Sheenah Nsubuga started this project — Jun 02, 2025 02:24 PM EDT

Leave feedback in the comments!

Log in or sign up for Devpost to join the conversation.