FinAI

• 

  NER Using gemini from Urdu document

Inspiration :

The inception of this project was motivated by a simple yet profound realization: financial services, despite their ubiquity, remain inaccessible to a significant portion of the global population. Newcomers, including immigrants and refugees, often find themselves on the peripheries of the financial system, hindered by barriers such as lack of documentation, credit history, and knowledge of local financial norms. Inspired by the potential of technology to bridge gaps, the project was born out of a desire to create a more inclusive financial world where everyone, regardless of their background, has access to financial services.

What it does :

Given the above problem, we are using Google Gemini to extract features such as name, address, date of birth, father's name, etc., in English from the identification document given by the individual in the regional/local language, for example, Hindi, Urdu, etc.

How we built it :

The model was engineered to perform named entity recognition from multilingual images by leveraging the Google technology stack, particularly Gemini, a multimodal platform.

Challenges we ran into :

Data quality and availability, model complexity, and compute requirements. Currently using the pre-trained model, a fine-tuned model with better datasets would greatly help.

Accomplishments that we're proud of :

Innovative use of technology, simplicity, and effectiveness can bring about building a financial inclusion model that has a societal impact.

What we learned :

SDG goals of UN - Goal 10: Reduce inequality within and among countries, Goal 17: Financial Inclusion Hands-on of the offerings from Google and interacting with Google Gemini and Gemma

What's next for FinAI:

Working further to fine-tune and add more features to the project. Expanding the project on Indigenous Languages and support for other less recognized languages.

Built With

• gcp
• gemini
• generativeai
• multimodal
• vertexai