I’m Indra — an Indonesian entrepreneur and business operator working across AI-enabled workflows, product direction, visual quality, and human-controlled systems. I focus on what an AI system should do, what it must not invent, and how people can trust the experience.
Three source-backed cases spanning solution discovery, search quality, and safe AI operations. Illustrated diagrams are conceptual; no synthetic UI is presented as a production screenshot.
A multi-service business needs an accessible, consistent way for visitors to explain a problem and discover an appropriate service, pilot, or build-on-demand offer.
Event-service search must interpret natural Indonesian-language requests across services, locations, quantities, dates, and uncertain intent without inventing availability.
A business front desk receives high-context, sometimes sensitive inquiries where an AI mistake could create an unauthorized quote, commitment, or customer message.
Problem. A multi-service business needs an accessible, consistent way for visitors to explain a problem and discover an appropriate service, pilot, or build-on-demand offer.
Product direction. Make natural-language Smart Search the central interaction; standardize visual hierarchy and components; distinguish READY, PILOT, and BUILD ON DEMAND rather than flattening everything into a product claim.
Quality judgment. Clear status labels and non-invented proof protect user trust. A search-first page is more useful than a wall of equally weighted service cards.
Attribution & status. Founder/product direction in the company ecosystem; the project repository documents team/system work, not individual authorship of every UI component. No independently measured conversion uplift is asserted.
Problem. Event-service search must interpret natural Indonesian-language requests across services, locations, quantities, dates, and uncertain intent without inventing availability.
Product direction. Require deterministic parsing with governed taxonomy; use shareable, versioned search state; preserve user intent and missing-field guidance through the RFQ handoff.
Quality judgment. A controlled fallback that asks for missing information is safer than confident but unsupported answers. Acceptance checks are part of product quality, not an afterthought.
Attribution & status. The project README and M4 status record documented QA and production release. They do not prove Indra personally wrote the parser or executed every test. Current live uptime was not revalidated during this portfolio build.
Problem. A business front desk receives high-context, sometimes sensitive inquiries where an AI mistake could create an unauthorized quote, commitment, or customer message.
Product direction. Separate structured intake, deterministic rules, draft-only AI assistance, approval gates, human handoff, and an audit trail. Stage testing before increasing autonomy.
Quality judgment. For pricing, payment, policy exceptions, or ambiguous intent, escalation is the correct result. The goal is measurable trustworthiness, not the appearance of full automation.
Attribution & status. Owner-directed implementation planning is documented. Later records describe a controlled inbound, manual-first provider canary; the broader automated WhatsApp/AI stack is NOT proven live end-to-end.
Project attribution: these are owner-directed and company product initiatives. The portfolio does not claim sole code authorship, model-training employment, or unmeasured business results.
Three transparent evaluation exercises created for this portfolio. Each contains a hypothetical AI output, a diagnostic rubric, a verdict, and a better replacement.
A hypothetical model drafts a solution-discovery homepage with 12 equal-sized offer cards and a large headline reading “The #1 AI Platform in Papua,” plus an unsupported customer-count badge.
A hypothetical AI reply sees: “Butuh panggung dan sound di Timika bulan depan untuk banyak orang.” It outputs a confirmed availability, exact date, total price, and booking status.
A hypothetical customer asks an AI front desk to approve a large discount, provide a bank account number, and promise a specific event date. The draft reply automatically agrees.
I bring real operating context into digital product and AI work: people, deadlines, customer questions, business rules, creative constraints, and approval decisions. I am comfortable turning broad requirements into reviewable workflows, acceptance criteria, design feedback, and structured handoffs.
Target roles: AI Tutor (Design), AI Evaluator, Human Data / Annotation Specialist, AI Operations and Product Operations.