← Selected work

Fintech product · Explainable AI · Data controls · 2024 to 2026

BRINK AI Funding Navigator

An award-winning fintech concept developed into a working, source-grounded product prototype for Indonesian MSMEs.

BRINKExplainable capital guidance for Indonesian MSMEsPrototype
Capital readiness72Operating well. Documentation still catching up.
Best route to investigate

KUR Working Capital

91% alignment based on productive use, operating history, and repayment preference.

Source checked28 June 2026
Rules decideSources stay visibleHuman verification remains required

Independent portfolio prototype. Guidance is illustrative and does not predict approval or replace provider assessment.

01

From winning concept to working product

The strategy case did not end with the pitch deck.

BRINK began as a fintech proposition for the Tech in Asia x Bank Rakyat Indonesia FutureMakers challenge, placing second among more than 940 participants. The original work combined user research, product strategy, financial modelling, and go-to-market design.

I later translated the concept into an interactive application that makes financing pathways easier to understand without presenting a black-box recommendation. The product now demonstrates how I move from commercial framing to interface design, decision logic, data controls, and deployable software.

2ndcompetition result
940+participants
4assessment dimensions
3pathways compared
02

Product logic

Separate eligibility, suitability, evidence, and explanation.

Structured assessment

Captures operating stage, use of funds, available financial records, and preferred financing obligation through a four-step flow.

Deterministic boundary

Explicit requirements and control states remain rule-based. Generated language is never positioned as the authority for eligibility.

Source-linked pathways

Each financing route keeps its rationale, caveat, institution, source URL, and verification date visible in the interface.

Explainable guidance

The coach layer turns structured results into plain-language reasoning and next actions while preserving uncertainty.

03

Technical implementation

A real interface, built as a product system.

The deployed prototype is intentionally front-end only. It demonstrates product logic, evidence design, quality controls, and interaction patterns without collecting identity documents, financial credentials, or user data.

ApplicationNext.js 16 · React 19 · TypeScript

App Router project with typed content models, reusable UI states, optimized local assets, and Vercel-ready production configuration.

Interaction stateAssessment · comparison · roadmap

Client-side state coordinates multi-step answers, selected financing routes, coach responses, task completion, and view transitions.

Evidence modelSource · freshness · caveat

Recommendation objects preserve traceability to primary sources and keep verification status beside the output rather than in hidden metadata.

Quality controlsIngest · normalize · validate · review

A dedicated control surface models completeness, validity, consistency, freshness, exception severity, and human-review states.

04

Designed control flow

Recommendations should be publishable only when their evidence is.

The quality console turns a product-content problem into an auditable data workflow. It is represented in the prototype as a production design, not as a claim that a live ingestion backend is already running.

  1. 01
    Ingest

    Official program, regulator, and market source records.

  2. 02
    Normalize

    Standard fields for route, provider, requirements, limits, and dates.

  3. 03
    Validate

    Completeness, validity, consistency, and freshness controls.

  4. 04
    Review

    Exceptions remain visible until a human resolves or blocks them.

  5. 05
    Publish

    Only reviewed evidence becomes recommendation content.

01Primary sources
02Pathway schema
03Rule controls
04Readiness logic
05Explanation layer
06Human action plan
05

Responsible AI boundary

The current prototype is honest about what is and is not automated.

Eligibility authorityDeterministic rules
Current coach responsesCurated and source-grounded
Live language modelNot connected
Personal financial dataNot collected
Final decisionHuman and provider-owned

A production version would add retrieval over versioned source records, citation enforcement, evaluation sets, prompt and output logging, and escalation when evidence is missing. The core boundary remains the same: AI may explain a controlled result, but it cannot silently create a rule.

06

What the work demonstrates

Product judgment that moves across finance, data, and implementation.

BRINK is the clearest expression of how I work: frame the institutional and user problem, turn ambiguity into explicit decision logic, make the evidence legible, and build the interface people actually use.

Fintech product strategy and user journey designFinancial pathway comparison and caveat designTyped front-end development and production deploymentData-quality workflow and human-review modellingExplainable AI boundaries and source-grounded communicationIndonesia-specific market and institutional context