Insights Hub – NUS FinTech Lab

FinTech Factory LunchTime Series – 9

We’re excited to host the next session of our FinTechFactory Lunchtime Series with

“Beyond the Hype, Scaling Social Agentic Oracles for High-Frequency Prediction Markets”

Date: 25 February, 2026 (Wednesday)
Time: 02:00 PM to 03:00 PM SGT
Presenters: Junwei (Zheng Junwei), Founder & CEO of Overheat

Join us for an inside look at Overheat, a social agentic oracle designed to revolutionize how we price information in real-time. While traditional prediction markets struggle with fragmented liquidity and slow resolution, Overheat leverages autonomous AI agents to research, verify, and resolve complex social and market trends on the Solana blockchain.

In this session, we will explore:

The Oracle Problem: Why traditional data feeds fail for subjective or high-frequency social events.
Agentic Intelligence: How multi-agent frameworks move from “data scraping” to “autonomous verification” to ensure market integrity.
Performance Moat: Leveraging sub-second finality to create a seamless, social-first forecasting experience.The Future of Information Markets: How Overheat transforms abstract public sentiment into a tradable, verifiable asset class.

Key Themes: AI Agents, Prediction Markets, Oracle Infrastructure, Solana Web3

Can’t join us in person? Join via Zoom

FinTech Factory LunchTime Series – 8

Insights Hub

Brownbag series recap

Date: 28 January, 2026 (Wednesday)
Time: 02:00 PM to 03:00 PM SGT
Presenters: Say Keong, Keith – NUS undergrad students

This session introduces Fundl’s Data Plugin ecosystem, a new approach to decentralized fundraising that transforms static crowdfunding pages into real-time, verifiable growth dashboards. We will explore how Fundl aggregates live signals from sources such as GitHub, Stripe, Google Analytics, social platforms, and market benchmarks to surface trustworthy indicators of momentum and execution. The discussion will cover the challenges of reducing information asymmetry in early-stage investing, the design of high-trust data plugins layered on top of decentralized fundraising and the strategic rationale behind a data-first model. Participants will gain an inside look at how Fundl enables backers to identify high-potential projects earlier, perform deeper due diligence, and participate in crowdfunding with venture-grade insight at mass-market scale.

Fundl’s Data Plugin ecosystem

Market Opportunity

Crowdfunding platforms represent a US$2.14 billion market (2024) dominated by incumbents like GoFundMe and Kickstarter. Fundl positions itself as a blockchain-powered alternative capable of disrupting the space by offering greater transparency, accessibility, and efficiency.

Key Features of Fundl

1. Decentralised fundraising

On-chain transparency showing how much funding is raised and how funds are used
Smart contracts enabling programmable accountability
Cross-border participation with reduced reliance on traditional banking systems
Open participation for both creators and backers

2. Benefits for project creators

Lower platform fees (2.5% vs ~5% on traditional platforms)
Open access without centralized gatekeeping
Streaming payments, where pledged funds are released over time through smart contracts

3. Benefits for funders

Transparent and trackable transactions on the blockchain
Refund mechanisms via majority voting
Early adopter benefits such as lifetime pricing or priority access
Real-time execution metrics (e.g., GitHub activity, revenue signals)

Credibility Through Verified Data

Fundl integrates external platforms such as Stripe, GitHub, and Google Analytics to create a system of record for project metrics. These integrations help:

Prevent inflated metrics
Provide objective benchmarks for comparing projects
Replace narrative fundraising with verifiable evidence of traction
Reduce perceived risk for backers

Strategic Feedback from Discussion

Participants suggested potential enhancements:

Introduce fixed donation tiers and flexible contribution models
Enable non-monetary contributions (e.g., coding, marketing) with rewards such as revenue share or project equity
Explore an always-on fundraising model structured in seasons or batches to maintain engagement

Fundl next steps:

1. Complete remaining integrations and prepare for mainnet deployment

2. Conduct targeted outreach to creators already active online or building publicly

3. Leverage startup ecosystems such as Product Hunt, accelerators, and university entrepreneurship programs

Key Takeaway

Fundl aims to redefine crowdfunding by combining blockchain transparency, programmable funding mechanisms, lower fees, and verified performance metrics, positioning itself as a credible next-generation funding platform.

FinTech Factory LunchTime Series – 7

Insights Hub

“Unbrowse Agentic Web”

Date: 19 November, 2025 (Wednesday)
Time: 02:00 PM to 03:00 PM SGT
Presenters: Lewis Tham – NUS undergrad student

This session introduces Unbrowse, a compatibility layer that enables autonomous agents to interact seamlessly with the open internet. We will explore the practical challenges encountered while building this system—from handling diverse web environments to ensuring reliability and safety at scale. The discussion will also highlight key lessons learned, design considerations, and emerging opportunities for agent-based automation. Participants will gain an inside look at how Unbrowse bridges the gap between agents and real-world web interfaces.

Problem

The pitch argues that the internet is currently fragmented and difficult for AI agents to use effectively:

Data and services are locked behind siloed platforms and restricted APIs.
AI agents often rely on GUI-based automation (screen scraping), which is slow and breaks when interfaces change.
User interactions generate valuable insights, but most of that value is captured by large corporations, not users.
These limitations prevent AI agents from scaling across the web.

Solution

Unbrowse proposes a network-level wrapper that converts web activity into structured “abilities” that agents can reuse:

Intercepts network packets instead of relying on UI automation.
Converts real user sessions into reusable agent tools.
Builds a universal index of abilities that agents can discover and combine.
Uses recommendation systems to match agent intentions with optimal execution paths.

How It Works

1. Capture & Upload – A browser extension records web sessions (credentials remain local).

2. Derive Abilities – Server processes sessions into reusable agent tools.

3. Discover & Run – Agents query the index and execute workflows via proxies.

4. Earn & Evolve – Users earn rewards when their contributed abilities are used

Key Innovations

Network-Level Precision: Uses packet-level data instead of GUI automation for higher reliability.

Collective Intelligence: Aggregates user behavior to produce broader insights and trends.

Unlimited Discoverability: Agents can dynamically find and combine tools.

User Monetization: Contributors earn rewards when their data-derived abilities are adopted.

Market Opportunity

AI agents market projected to grow from $5–7B in 2025 to ~$42B by 2030.
Year-1 goals:
- 10,000 users
- 50,000 indexed abilities
- ~$1M ARR from SaaS plus ~$500K in token fees.

Vision

Unbrowse positions itself as infrastructure for the “agentic internet”, enabling AI agents to interact with the web seamlessly while allowing users to earn value from their interactions.

Congratulations to the NUS FinTech Society Team at ETH Rome 2025!

Insights Hub

Congratulations to our FTS representatives for their incredible showing at ETH Rome 2025! From October 17-19, our members took away 4 prizes and demonstrated immense focus and creativity, reflecting the society’s deep talent and offering a glimpse into the future of blockchain.

Let’s hear from the team about their experience.

ETH Rome 2025

Billed as the “leading ETHalian Hackathon crafted by builders, for builders,” ETH Rome 2025 invited developers to the vibrant Talent Garden Roma. The event focused on Web3’s most pressing challenges, with key tracks in AI, DeFi, and Privacy, and featured a prize pool exceeding $60,000 USD.

The Process

In the midst of a beautiful Roman autumn, five representatives from NUS FinTech Society hacked tirelessly, all as solo developers. The hackathon officially kickstarted on the evening of October 17th, but our efforts began long before the flight to Rome. Once the tracks were announced, we started brainstorming ideas and reaching out to mentors and sponsors, including teams from Civic, iExec, ENS, and Base, to explore the problem statements and technical details.

The hacking environment at Talent Garden Roma was amazing. It was a fantastic opportunity to meet with fellow hackers and friends in the Web3 ecosystem. We even managed to reunite with hackers we met in previous hackathons like ETHOxford! The venue was beaming with life day and night, and it seemed people did not sleep at all. The organizing team also kept us fueled with amazing Roman food, from pizza and lasagna to pasta and, of course, gelato!

The final night before the deadline was a true test of endurance. We barely slept, pushing to get our products fully functional. Sustained by caffeine and pure adrenaline, we battled broken codes and last-minute bugs. After more than 24 hours straight without sleep, we finally shipped our projects right at the 10 a.m. deadline.

After the intense hacking, we celebrated by walking into the mesmerizing, dreamy sunset toward the Colosseum and ended the night with the best carbonara ever. Hearsay Jeff finished his first plate in 2 minutes and ordered another one straightaway.

The crowd in opening ceremony

Locking in on the final night

Wei Rong and Kevin discussing the projects

Featuring Say Keong in a celebratory mood

The Outcome

These are the 5 projects built by our solo hackers.

Smart Nexus

Civic 1st Place

An AI-powered Web3 chatbot that identifies smart money token movements through Nansen AI MCP integration and executes instant token swaps via 1inch API, all secured through Civic Auth Web3’s embedded wallet for seamless transaction signing.

GitHub Link: https://github.com/jefflcy/ethrome-2025

Hackathon Submission: https://taikai.network/ethrome/hackathons/2025/projects/cmgx6tmzh02e412kk0klkxcds/idea

ExecSwap

iExec 3rd Place

A secure, privacy-preserving vault that lets users swap ERC20 tokens via cryptographic commitments and withdraw safely while tracking balances off-chain.

GitHub Link: https://github.com/PohSayKeong/ExecSwap

Hackathon Submission: https://taikai.network/ethrome/hackathons/2025/projects/cmgwj3q5z01vn12kk2hwdveye/idea

Munus

Civic 2nd Place

Chat-native job marketplace where AI agents coordinate workflow and Base blockchain secures trustless payments. Post jobs in XMTP group chats, authenticate with Civic, deliver work with attestation and get paid instantly on Base L2.

GitHub Link: https://github.com/Nimastic/Munus

Hackathon Submission: https://taikai.network/ethrome/hackathons/2025/projects/cmgxalmsy01zc8697b7iyvwbi/idea

Trick or TrETH

ENS Pool Prize

On-chain trick or treating where you design your own costume or judge others who visited you! A fun and intuitive way to celebrate this Halloween on crypto.

GitHub Link: https://github.com/wr1159/ethrome-2025

Hackathon Submission: https://taikai.network/ethrome/hackathons/2025/projects/cmgxal4br01lz1nb96rqwxppk/idea

BasePay MiniApp

A mini app on the Base app that allows merchants and users to leverage existing QR code infrastructure to deal and transact in stablecoins.

GitHub Link: https://github.com/lim-4158/BasePay-MiniApp

Hackathon Submission: https://taikai.network/ethrome/hackathons/2025/projects/cmgwujcpd01ms8697evyvddxq/idea

Quotes

What stood out during this trip was how competitive the Web3 development scene is in Italy, as there was a month-long "hack village" prior to the hackathon where hackers stayed together and attended workshops held there, and then collectively worked on some projects to hone their skills. As such, we faced immense competition from locals that were already well in the zone and ready to hack over the weekend. Luckily, as there were various bounties to compete for, we shortlisted them accordingly and innovated extensively on bold ideas that could stand out and secure prizes with the new tech stacks from the bounty sponsors.

Jefferson

My experience at ETHRome was nothing short of amazing. Across the 3 days, i had the opportunity to network and attend workshops where i gained a better understanding of the sponsor's underlying blockchain technology. I iterated on my project after receiving feedback, and line by line, built a project that the sponsor felt has real business use case. As the popular saying goes, Rome wasn’t built in a day, but Munus was built in 3.

Jeriel

Mamamia!

Kevin

Participating in EthRome has expanded my knowledge in the field of blockchain and especially DeFi. Through experimentation and research, I built a protocol to facilitate trading on the blockchain directly. I have also learnt more about how trusted execution environments can be integrated into Dapps to create new and innovative privacy solutions.

Say Keong

ETHRome opened up my eyes to how competitors used AI in assisting their hackathon submission effectively. There I realised that since execution of most projects no longer becomes a problem with the help of AI, we must all learn to become more critical and creative thinkers to come up with standout ideas or build an even more technically challenging solution.

Wei Rong

Gallery

FinTech Factory LunchTime Series – 6

Insights Hub

“fIDent : Credit You Can Carry”

Date: 31 October 2025 (Friday)
Time: 03:00 PM to 04:00 PM SGT
Presenters: Aniroodh Chaudhary, Arnav Jhajharia – NUS undergrad students

This session explores the emergence of fIDent as a reimagined framework for financial identity in the global gig economy. Amid rising demands for credit inclusion, fragmented KYC systems, and opaque data portability, fIDent proposes a unified, blockchain-anchored “financial passport” that travels with the worker across platforms and borders. Through a synthesis of behavioral analytics, open-finance integrations, and verifiable identity layers, the study examines how decentralized credentials can bridge the trust gap between informal earners and formal lenders. It probes key tensions—between privacy and transparency, decentralization and compliance, innovation and regulation—and situates fIDent within the broader evolution of digital identity infrastructure. Ultimately, the work envisions a more interoperable and equitable financial architecture where reputation, not paperwork, powers access.

The Problem

Globally, a huge portion of workers cannot access formal credit.

60% of the global workforce is informal

1+ billion gig and informal workers have no formal credit history

1 in 4 adults borrow from informal lenders

<20% of gig workers have access to formal credit

Key reasons:

No formal credit profile

Financial data scattered across platforms

Lenders cannot properly assess risk

Workers rely on exploitative lending sources

Market Opportunity

Despite lacking credit access, gig workers show strong financial behavior:

76% have bank accounts
50% save for emergencies
72% send remittances regularly
97% repayment rate in microfinance
70% accept digital payments

➡️ Conclusion: They are creditworthy but invisible to lenders.

The Solution: fIDent

fIDent builds a decentralized “financial passport” for gig workers.

It converts digital financial activity into a portable credit identity.

Key idea:

Turn “invisible workers” → “credible borrowers.”

The system uses:

Digital transactions
Behavioral patterns
On-chain credentials
AI-based credit scoring

How the Platform Works?

Step 1: Identity Creation

Users authenticate via idOS
KYC through SumSub
Digital identity created using XRP DID protocol

Step 2: Connect Data

Users link financial/work platforms via:

Argyle
Plaid

Step 3: Loan Application

Borrowers apply through a lender marketplace
Lenders match based on the fIDent score

The Scoring Engine

fIDent combines financial + behavioral data to assess creditworthiness.

Financial data

Balance history
Cash flow
Loan/deposit patterns

Behavioral data

Gig consistency
App ratings
Device usage patterns

AI processes this data to generate a transparent credit score.

Explainability Layer

Unlike traditional credit scores, fIDent provides transparency. We uses the three tools:

1. User Dashboard

Shows income, spending, and financial health
Gives tips to improve credit score

2. Lender Analysis Report

Detailed explanation behind credit decisions

3. Post-Disbursal Monitoring

Detects early risk signals to prevent defaults

Value for Users

Users receive a portable financial identity that unlocks:

Affordable loans

Job opportunities

Rental agreements

Insurance and financial services

They also get a financial dashboard to track spending, savings, and income.

Value for Lenders

Lenders gain:

Dynamic underwriting reports
Behavioral credit analysis
Blockchain-based loan escrow
Real-time borrower monitoring

Benefits:

Reduced default risk
More loan approvals
Access to a large new borrower segment

Market Size

TAM: $500B
SAM: ~$50B (digitally active but credit-invisible users)
SOM: ~$2.5B

Target users:

Gig workers
Freelancers
Informal workers
Migrant laborers

Business Model

Revenue streams:

B2B API
- Credit scoring engine for lenders
Targeted marketing
- Access to underserved gig worker segments
Co-selling financial products
- Loans, insurance, savings products

Go-to-Market Strategy

Growth strategy includes:

Word-of-mouth in gig communities
Partnerships with gig platforms
Pilot programs with lenders
Incentivized referrals

Competitive Advantage

fIDent differentiates itself by:

Building financial identity + credit scoring
Using behavioral + financial data
Offering explainable AI
Creating a global credit infrastructure

Financial Projections

Growth plan:

Year	Users	Revenue
Year 1	20K	$0.3M
Year 2	100K	$1.7M
Year 3	700K	$11.1M
Year 4	1.5M	$28.5M

Profit expected starting Year 3.

In summary:
fIDent is a fintech platform that builds portable, AI-powered financial identities for gig and informal workers, enabling them to access credit while helping lenders assess risk using financial and behavioral data.

FinTech Factory LunchTime Series – 5

Insights Hub

“Beyond Borders: Reimagining CBDCs, Stablecoins and RTGS for Cross-Border Settlement”

Date: 26 September, 2025 (Friday)
Time: 02:00 PM to 03:00 PM SGT
Presenter: Boonthicha Sae-Jia, NUS PhD Student

1. Inefficiencies in Current Cross-Border Payment Systems

The existing cross-border payment system is often slow, expensive, and inefficient due to its reliance on traditional correspondent banking networks. Transactions typically involve multiple intermediaries, leading to settlement times of two to three days and transaction fees of around 1.5%. Payments between countries frequently pass through major currency clearinghouses such as the US dollar or euro systems, which adds further delays and operational costs. These structural limitations create scalability challenges and increasing dissatisfaction among businesses that require faster and more cost-efficient international transactions.

2. Global Efforts to Modernize Payment Infrastructure

To address these challenges, international initiatives such as the G20 roadmap aim to modernize cross-border payments by improving speed, access, transparency, and operating hours. Proposed measures include expanding access to real-time gross settlement (RTGS) systems to non-bank participants and implementing the ISO 20022 messaging standard to improve interoperability and data consistency. Although many RTGS systems have adopted the new messaging standard, extending operating hours to a full 24/7 model remains difficult due to legacy infrastructure and operational constraints.

3. Role of Blockchain, Stablecoins, and CBDCs

Emerging technologies such as blockchain, stablecoins, and central bank digital currencies (CBDCs) are being explored as alternatives to improve payment efficiency. Stablecoins provide fast settlement and lower transaction costs but face significant regulatory and transparency concerns. Their decentralized nature makes ownership verification, auditing, and sanction compliance more difficult. In contrast, wholesale CBDCs offer stronger transparency through regulatory oversight, but their integration with existing systems can limit cost efficiency and increase operational complexity.

4. Operational and Integration Challenges

Despite their potential benefits, both stablecoins and CBDCs face integration challenges with existing financial infrastructure. RTGS systems still rely on manual processes and limited operating hours, which restrict real-time settlement and reduce efficiency. Wholesale CBDCs must also interact with legacy systems such as RTGS, which can offset some of the advantages offered by blockchain-based technologies. As a result, modernizing cross-border payment systems requires upgrading legacy platforms, improving interoperability, and gradually integrating new technologies while maintaining regulatory compliance and financial stability.

FinTech Factory LunchTime Series – 4

Insights Hub

We’re excited to host the next session of our FinTechFactory Lunchtime Series with

“Reconfiguring Trust in Centralized Systems: Evidence from Permissioned Blockchains in the Payment and Settlement Industry”

Date: 10 September, 2025 (Wednesday)
Time: 02:00 PM to 03:00 PM SGT
Presenter: Kenneth See, NUS PhD Student

This session tackles the core paradox of blockchain in centralized, high-value financial systems: if blockchain is “trustless”, why do we see prolonged adoption even in centralized systems—where institutional trust is indispensable? Drawing on a qualitative case in payments and settlements, the work reframes blockchain as a hybrid socio-technical infrastructure that redistributes, rather than replaces, trust. We propose a recursive conceptual model that explains how organizational governance, regulatory constraints, and protocol design co-evolve to reconfigure trust, paving the way for continued adoption, over time. This study contributes a theoretically grounded, policy-relevant lens for understanding sustained blockchain use in regulated markets, moving beyond simple “transfer of trust” narratives to a dynamic account of trust reconfiguration.

Can’t join us in person? Join via Zoom

#PaymentSystems #FinTechResearch #NUSFinTech #DesignScience #ComplexSystems

NUS FinTech Lab Interns Take XRPL Innovation Global

Insights Hub

At NUS FinTech Lab, we empower students to explore, experiment, and innovate. Recently, our interns Ong Si Hui (Ariel) and Do Trong Uy (Wyn) represented the Lab at major XRPL programs in New York and Paris, gaining hands-on blockchain experience and global perspectives.

Ariel – XRPL Student Builder Residency 2.0, New York

Three days of travel from Singapore to New York between 13 Aug and 15 Aug, 2025 were just the beginning of Ariel’s journey at the XRPL Student Builder Residency 2.0.

Over four weeks, she transformed an idea into RePawn, the first gold-pawn shop on the XRP Ledger. The project allows gold to be pledged as collateral, automates compliance with XRPL Hooks, and manages risk on-chain.

Her demo at Demo Day showcased the potential of blockchain to traditional finance and highlighted the collaborative innovation of student builders worldwide.

“The late-night debugging, constant feedback, and collaborative spirit made every challenge worth it.”

Ong Si Hui (Ariel)

Wyn – XRPL Core Dev Bootcamp, Paris

Across the world in Paris, Wyn dove into the XRPL Core Dev Bootcamp. Over two intense weeks from 11 July to 25 July, 2025, he explored the inner workings of the XRP Ledger—running validator nodes, understanding consensus mechanisms, and navigating ledger structures.

Beyond technical skills, the bootcamp offered inspiration. “Meeting cryptographers and engineers who are shaping the future of blockchain was eye-opening,” Wyn says. Collaborative projects and mentorship helped him see not just how the ledger works, but how innovation happens—through curiosity, persistence, and teamwork.

“Meeting cryptographers and engineers shaping the future of blockchain was eye-opening. It reinforced the value of curiosity and continuous learning.”

Do Trong Uy (Wyn)

Learning, Exploring, Innovating

NUS FinTech Lab strives to offer opportunities for our interns to explore, experiment, and grow in global settings. We continue to empower our students to push boundaries, turn ideas into reality, and shape the future of FinTech.

FinTech Factory LunchTime Series – 3

Insights Hub

BrownBag Series Recap

Date: 31 July, 2025

Speaker: Alya R Prastiti, Intern @ NUS FinTech Lab

This brownbag explored the complexities of Singapore’s Option-to-Purchase (OTP) workflow for residential property transactions, shedding light on the inefficiencies that persist. The discussion revealed how manual hand-offs, fragmented compliance checks, and legal constraints hinder efficiency and transparency, while also outlining pathways for digital transformation.

Mapping the Gaps: Understanding and Redesigning Property Transaction Workflows in Singapore

Understanding the Problem

The current OTP process involves multiple stakeholders—property agents, conveyancing lawyers, mortgage officers, and government agencies such as the Singapore Land Authority (SLA) and the Inland Revenue Authority of Singapore (IRAS). Despite the availability of digital tools, the workflow remains heavily reliant on manual steps. Lawyers, for instance, are required to conduct searches across various portals:

INLIS for title checks,
eLitigation for litigation history,
MinLaw’s e-service portal for bankruptcy searches, and
STARS eLodgment for lodgments.

This fragmentation creates delays and additional costs, often forcing practitioners to turn to third-party services. Moreover, the continued requirement for wet-ink signatures—mandated under current regulations—prevents the adoption of more efficient, paperless alternatives.

Challenges in the Current Workflow

The speaker summarized several structural bottlenecks:

Fragmented portals leading to duplicated effort.
Absence of real-time notifications, slowing the flow of critical information.
Mandatory wet-ink signatures, which prevent full digital adoption.
Heavy compliance burdens, particularly AML/CFT checks and the retention of records for five years.

These challenges collectively limit both efficiency and transparency, raising questions about the long-term sustainability of the current system.

Research Questions Raised

The session framed the inquiry around three guiding questions:

What pain points most delay or compromise compliance and transparency?
Which digital capabilities do practitioners most value for efficiency?
How can digital tools be designed to integrate seamlessly with Singapore’s legal framework?

These questions served as the basis for stakeholder interviews and practitioner-led design exercises, aimed at identifying where digital interventions could have the most meaningful impact.

Towards a Digital Solution

To address these gaps, the session proposed a two-tier digital solution:

Integrated Due Diligence Platform
- A unified interface consolidating all searches and lodgments.
- Embedded KYC/AML checks and automated compliance workflows.
Real-Time Collaboration Dashboard
- A secure environment enabling property agents, lawyers, and regulators to coordinate seamlessly.
- Immediate updates on progress and compliance status.

This dual approach aims to streamline workflows while ensuring regulatory compliance.

Future Research Directions

Looking ahead, the next phase of work must go beyond digitizing existing workflows to developing a blockchain infrastructure blueprint that complies with Singapore’s property laws while supporting both traditional sales and emerging fractional ownership models.

Key research areas identified include:

Fractional Ownership – Investigating how to legally structure, govern, and record shared ownership rights on-chain in a manner consistent with Singapore’s property regulations.
Legal Entity Requirements – Exploring compliance frameworks for tokenized property ownership involving companies, trusts, or funds, ensuring alignment with KYC/AML standards and beneficial ownership rules.
Dissolution & Re-Sale Processes – Designing enforceable and technically robust mechanisms for fractional stake resale, buyouts, or dissolution, including smart contract–driven protocols that map onto existing conveyancing practices.

Key Takeaway

The session underscored that the future of property transactions in Singapore will hinge on bridging law and technology. While digitization can streamline existing workflows, the transformative leap lies in creating a blockchain-based infrastructure that enables secure, compliant, and legally recognized property sales—including fractional ownership models. Progress in this space will depend on targeted research into ownership structures, and legal entity compliance, paving the way for a tokenization-ready property market.

FinTech Factory LunchTime Series – 2

Insights Hub

BrownBag Series Recap

Date: 24 June, 2025
Speaker: Ryo Seah, Intern @ NUS FinTechLab

This BrownBag session focused on automating donor due diligence for philanthropic organizations, presenting a proof-of-concept OSINT pipeline. The discussion examined technical challenges including LLM hallucinations and clustering thresholds while highlighting solutions such as hierarchical clustering, source reliability scoring, and integration with CRM systems for scalable risk management.

An Automated Pipeline for Generating Open Source Intelligence Dossiers for Donor Vetting

Understanding the Problem

Philanthropic organizations face a complex challenge in ensuring comprehensive donor due diligence while managing reputational risks. Current workflows are often manual, fragmented, and time-consuming, requiring staffs to piece together unstructured data from scattered sources. To address these inefficiencies, the lab explored how intelligent automation—combining web scraping, large language models, and clustering techniques—can streamline the due diligence process and improve the accuracy and scalability of donor profiling.

Key Challenges

Two persistent challenges stood out:

LLM Hallucinations – While LLMs are effective at structuring scraped data, they occasionally produce unsupported claims.
- Solution: Establishing a “chain of trust” by tying each fact directly to its original source.
Ambiguous Profiles – Similar or identical names often led to merged donor records.
- Solution: Agglomerative clustering with semantic embeddings flagged potential overlaps for human review.

Other hurdles included API rate limits, blocked scraping on certain sites, and the scalability of processing larger name lists.

Technical Implementation

The demonstrated pipeline (Stage 1: OSINT Automation) follows a three-step workflow:

Search: SerpAPI for Google results, Playwright for scraping dynamic content.
Aggregate: LLMs structure data into clean key-value pairs, removing noise.
Cluster: Embeddings identify duplicate or ambiguous profiles, generating YAML outputs for PDF dossiers.

This approach ensures traceability by linking every fact to its source, significantly reducing hallucination risks.

Research Questions & Discussion

The BrownBag session not only outlined the proof-of-concept pipeline but also raised critical questions from the internal audience, reflecting both technical and strategic considerations:

How can the system be made more user-friendly, perhaps through a GUI interface with simple input fields rather than relying on command-line operations?
What statistical approaches (e.g., elbow method) could refine clustering thresholds and improve the disambiguation of donor profiles?
Can Named Entity Recognition (NER) be integrated to automatically filter out irrelevant paragraphs during web scraping?
How should we design a source reliability scoring system that prioritizes information from more authoritative and trustworthy domains over less reliable sources?
Would a saturation-based link collection approach, instead of a fixed link limit, improve coverage and accuracy of profiles?

Key Takeaways

Proof of Concept: Successfully processed 5 names per batch with 70–80% scrape success, producing structured and verifiable YAML/PDF outputs.
Audience Feedback: Raised crucial ideas for GUI development, clustering refinements, NER, and open-source LLM exploration.
Future Roadmap: Enhancements include source reliability scoring, fuzzy matching across sources, and integration of risk indicators.
Strategic Outlook: Beyond automation, the project aims to build a full Intelligence & Risk Engine to support long-term donor due diligence.

The session closed with a shared recognition: while automation cannot replace human oversight, it can significantly reduce repetitive effort, improve accuracy, and enable staff to focus on high-value decision-making.