Few topics have hogged the headlines like Artificial Intelligence (AI) has all year. The recent development of Large Language Models (LLMs) has given AI the uncanny ability to imitate human intelligence – and for Funding Societies’ crack team of engineers, this is an opportunity that’s impossible to pass up.
Since 2018, Funding Societies has held an Annual Funding Societies | Modalku (FSMK) Hackathon: a competitive and collaborative event that divides all employees into teams, which then compete against each other to solve problems and build solutions based on a theme.
Last year’s winner, for example, built an “AV Assistant” that helped applicants solve registration problems without navigating out of the page for instructions.
Table of Contents
AI solutions on the agenda
This year’s theme focuses on projects that use AI, specifically Generative AI based on LLMs, for our Payment and Lending products. These models can write code, create financial plans, create ads, and summarise documents.
Sixteen teams signed up to participate in this year’s Hackathon, with cross-border teams from Singapore, Malaysia, Indonesia, Thailand, and Vietnam competing to produce the most effective AI-based solution in the span of a few days.
Everyone put their heads together, built innovative solutions, and explored a broad range of potential solutions that generative AI can enable for ordinary businesses. The top five teams outdid themselves: presenting solutions that creatively leveraged LLMs to reduce transaction frictions, reduce costs, and improve the way Funding Societies helps our customers.
5th Place: LLM-based document parser
Meet the team: Som, Zeeshan, Ravi and Balaji of “MAITrix”
Their AI-based solution: “MAITrix” built an in-house bank statement parser that could scan PDF and JPG/PNG bank transactions, and extract and store banking data without human intervention.
The solution used OCR (Optical Character Recognition) LLMs to extract text from the PDFs and scanned images; and OpenAI GPT-3.5 Turbo to gather bank transaction data from the extracted text, and convert it into JavaScript Object Notation (JSON) format.
Extracted and formatted data is stored in a relational database, ensuring easy accessibility and data integrity.
Outcomes and insights: The solution combined two different AI approaches: OCR LLMs extracted text from unstructured documents, while machine learning models classified and extracted specific information from the data.
By combining the two, the new parser offered greater levels of automation, accuracy, and efficiency compared to the costly, outsourced solution. With widespread implementation, the team believes their solution can reduce document parsing costs by 50%.
4th place: AI-based customer support service
Meet the team: Melissa, Vikram, Jiah Hooi and Cristel of “Charlie and Angels”
Their AI-based solution: “Charlie and Angels” explored ways to leverage AI for instant communication, personalised support, and 24/7 availability, ultimately enhancing the overall customer experience.
They hoped to “revolutionise customer engagement” through generative AI: transforming customer interactions through an AI chatbot that could provide accurate and efficient responses to user inquiries.
Outcomes and insights: Once the solution can be integrated into Funding Societies’ customer engagement strategies, the team hopes to increase monthly active users with the use of AI-enabled instant communication, personalised support, and round-the-clock availability.
3rd Place: AI-powered ad campaign generator
Meet the team: Binay, Tze Nam, Pinkie and Debi of “Marketing Maverick”
Their AI-based solution: Inspired by ChatGPT, “Marketing Maverick” presented a project that leant on their marketing chops: a process for quickly churning out compelling advertising creatives and campaigns based on the Midjourney generative AI tool.
Outcomes and insights: The marketing team is ready to develop a pilot project based on their Hackathon entry. “We will generate a set of creatives based on a few themes, and try them out in our ad campaigns,” the team explained in their presentation. “The themes have included industries, types of financing, countries, etc.”
Marketing Maverick also plans to “templatise” the AI prompts used in their solution. “Templating the prompts ensures all the stakeholders can generate the creatives in the shortest time,” they explain. We can use these templates to quickly create messages that respond to evolving issues and relevant trends for our customers across the region.
2nd Place: Recommendation engine
Meet the team: Agus, Trong, Pan Min and Rajesh of “Autom8ers”
Their AI-based solution: When onboarding borrowers, FMSK uses a basic recommendation system based only on certain info the borrower provides in their onboarding form. The “Autom8ers” want to build a more comprehensive and nuanced recommendation system. Their solution is a rule-based recommendation engine that automatically suggests products for a borrower based on all data that was used in the origination and underwriting process. AI converts the information in the documents into structured data that the engine can query.
Outcomes and insights: With AI, FMSK staff don’t have to manually pore every single document about a borrower. The “Autom8ers” used AI to understand and parse relevant data from the documents. The data is fed to a rule engine, which recommends products for a borrower based on certain criteria.
1st Place: Churn prediction model
Meet the team: Eu Jin, Evan, Jangam and Siddhant of “ChurnWhisperer”
Their AI-based solution: Defaulting users and “churn” (the rate at which customers stop using a service) are a challenge for the Elevate Credit Line product. The aptly-named “ChurnWhisperer” team sought to use AI, prediction models, and machine learning to help business users within FSMK identify default and churn signals. The model helps users spot customers with traits and behavioural patterns that match those signals.
Outcomes and insights: The team prioritised processes that helped non-technical business users to understand the analytical models and methods. They accomplished this with explainable AI libraries like Shapley Additive Explanations (SHAP) to bring the interpretation of the model predictions down to the users’ level.
Conclusion
Funding Societies’ Hackathons are a major event for our engineers and the rest of the company: as one engineer put it, “It’s like an annual festival for the tech team!”
It’s only a fiesta on the surface: the FSMK Hackathons offer an annual opportunity for engineers to “get together, innovate and take leaps to build the future,” as our ex Chief Technology Officer Ishan Agrawal explains.
Every year, FSMK Hackathons have sharpened our staff’s innovation and technical skill sets, while bonding employees across departments and regional offices with a sense of shared accomplishment.
Will any of the winning teams’ solutions move from planning to reality in the immediate future? Watch this space!
And if these projects excite you, check out our career opportunities page to find a fulfilling engineering role spent building and hacking — not just coding.
Not an engineer but interested in serving SMEs through FinTech? Check out other non-engineering opportunities at FMSK.
Disclaimer: The information provided to you in this blog post is intended only for general information purposes only and does not constitute legal or other professional advice on any subject matter. The materials and the information provided are not intended to be and do not constitute an advertisement or solicitation. In no event will Funding Societies be liable to any party for any direct, indirect, incidental, special, consequential or punitive damages for use of such information by you or any unauthorised third party.
- Pioneer to Leader: Meet Vikas Jain, FSMK’s First Employee - September 26, 2023
- Harvesting Opportunities: Innovations in Financing Capital-intensive Startups - August 22, 2023
- Enhancing Business Cash Flow through Accounts Receivable Financing - August 17, 2023