Boost Capital set out to modernize its Southeast Asian microfinance operations by replacing manual, spreadsheet-based workflows with an automated digital lending platform. Elementica engineered the backend infrastructure, data pipelines, and conversational UI, migrating years of fragile customer records into a secure relational database and building a Facebook Messenger bot that autonomously screens applicants, evaluates risk, and routes loans to partner financial institutions.
Fintech/Microfinance/Digital Banking
Client context
Boost Capital operates in Southeast Asia, providing micro-loans to small businesses and individuals underserved by traditional banking. To scale quickly in its early stages, the company relied on lightweight, low-cost tooling. Over time, this grew into a complex web of dozens of disconnected Google Sheets tracking customer verification docs, credit scores, active principal balances, and repayment schedules.
As transaction volume grew, human loan officers spent hundreds of hours manually cross-referencing files, verifying identity documents, and copying payment receipts. This manual setup created high operational costs, introduced frequent data-entry errors, and stretched loan processing times from minutes to days. Scaling the business meant linearly expanding human headcount — an unsustainable model for high-volume microfinance.
The challenge
Transitioning an active financial operation to an automated system required solving three interconnected engineering and product challenges:
- Legacy Data Integrity: Years of historical financial records sat in inconsistent formats, with missing fields, duplicate borrower entries, and unstandardized payment logs. This data had to be cleansed and unified without interrupting ongoing daily loan collections.
- Low-Friction Borrower UX: Target borrowers primarily accessed the internet via mobile social channels. The onboarding process — including document collection (KYC), personal details intake, and loan terms signing — had to occur entirely within Facebook Messenger, maintaining high completion rates without human assistance.
- Dynamic Credit Routing: Different microfinance institutions (MFIs) operate under distinct risk tolerances and lending criteria. The platform needed to evaluate applicant payloads in real time and dynamically match them to the right lender.
Elementica's role
Elementica acted as the full-stack technical lead. We designed the architecture from scratch, wrote custom migration routines, integrated third-party lending APIs, and built the core decisioning backend that orchestrates data flow between Facebook Messenger, our custom database, and external microfinance partners.
The solution
Zero-Downtime Data Migration Engine
We developed automated ETL (Extract, Transform, Load) parsing scripts in Node.js to clean, normalize, and reconcile scattered spreadsheet data. The script cross-referenced phone numbers, national IDs, and transaction histories to resolve duplicate profiles, backfilling missing metadata before pushing clean records into a relational PostgreSQL database. The migration ran in background stages, ensuring 100% data retention while the lending desk remained online.
Messenger-Native Conversational Intake
Instead of forcing users to download an app or visit an external site, we brought the entire lending branch into Facebook Messenger. We engineered step-by-step conversational guardrails that guide applicants through submitting national ID photos, business details, and income statements. The system validates uploaded images and data inputs instantly, returning immediate feedback if an image is unreadable or a field is malformed.
Multi-Lender Automated Credit Decisioning
Behind the chat interface sits a rules-based credit engine connected to three separate MFI platforms via custom API wrappers. When a user completes the application, the backend evaluates credit scoring rules in real time. If an applicant falls outside the eligibility criteria for Lender A, the engine automatically restructures and re-routes the profile to Lender B or C, turning potential drop-offs into approved loans.
Engineering decisions that mattered
Hybrid Frontend Architecture (ManyChat API + NestJS Backend)
Rather than allocating engineering resources to rebuild messaging infrastructure from scratch, we used ManyChat as a lightweight presentation layer while routing all business logic, validation, state management, and MFI communication through a custom NestJS service. This decoupled design saved months of frontend engineering while keeping critical credit logic in a secure, proprietary backend.
Idempotent Event Handling for Unstable Networks
Mobile connections in targeted rural regions are frequently spotty. To prevent double-disbursements or duplicated loan creation caused by dropped connections or repeated button taps, we enforced strict idempotency across all webhook handlers and API endpoints using unique transaction hashes.
Results
Boost Capital shifted from a slow, manual desk to an automated microfinance engine. The new platform processes loan applications in under 5 minutes, has autonomously financed over 40,000 borrowers, and saved over $60,000 in operational costs within 15 months — allowing the company to scale transaction volumes exponentially with a lean engineering team.
Product evolution
Following the core pipeline deployment, the product roadmap includes expanding conversational flows to WhatsApp and Telegram, integrating automated payment gateway links for automated debt collection, and training machine learning models on conversational behavior to further optimize credit risk scoring.
