Pronto - Sell Project

Pronto is a finished, shipping Android app: a work buddy that does the paperwork. The user types what they need in a chat ("make me a 10-slide deck on photosynthesis", "turn this photo of my notes into study cards", "turn this lecture recording into notes"), taps a skill card, fills a short form, and gets a real file back: .pptx, .docx, .pdf, .xlsx, a mind map, a transcript. It is completely free to use. No subscription, no card: users earn credits by watching rewarded video ads, and each skill is priced from its real AI cost. Live on Google Play (145 countries) and Uptodown, both on the current version 1.5.5. WHO IT'S FOR Students and junior office workers in emerging markets (Indonesia, Brazil and India first) who do all their work on a mid-range phone: no laptop, no international card, but a deadline and a file to hand in. Chat-only AI apps give them text and a paywall. Pronto gives them the finished deliverable, in their language, formatted to their local standard, for a few ads. HOW IT MAKES MONEY • One rewarded ad = 10 credits. New users get welcome credits, so the first tasks are free before anyone sees an ad. • Every skill is priced from measured token usage. The app holds a P90 amount, settles to the real cost when the task finishes and refunds the difference; users see "typically 6 credits, holds 27" up front and what they actually paid after. • The credit price is derived from rewarded-ad eCPM and skill prices are recomputed from the LLM provider's rate card, so ad revenue is designed to cover token cost with margin from the first user. A daily margin guard checks real settlements against provider rates. • Unit economics, roughly: skills are priced at about 2.5× measured token cost, so a typical slide outline (6 credits) costs around $0.0006 in tokens, and one completed rewarded view funds 10 credits. Break-even is about $1 eCPM, well below typical rewarded-video rates in Indonesia, Brazil and India. Per-task margin is positive from there up; what the product needs is volume, not a better rate. • Rewards are granted only after server-side verification (AdMob SSV, replay-safe); Play Integrity is wired in. Credits cannot be forged from the client. • Free where it earns trust: outlines are free until the user confirms them, regeneration is free, format conversion needs no AI at all. THE 34 SKILLS, AND WHY THEY ARE SHAPED THIS WAY A skill is a declarative card: an input schema (the form renders itself), a pricing mode, a prompt pipeline and an output renderer that produces a real file. Add a card and the chat router, quote card, form, pricing, store page and daily report all pick it up. Seven groups: • Presentation: PPT outline → full deck with speaker notes, speaker script, presentation bundle (deck + cover page in one job) • Documents: resume PDF, cover letter, ATS resume match, report DOCX, document summary, paragraph-by-paragraph translation, academic cover page • Academic, localized to real exam systems: ABNT formatting and references (Brazil), ENEM essay grading on the five official competencies, FUVEST/UNICAMP essay rubrics, LPDP scholarship essay planner and laporan praktikum lab reports (Indonesia), UPSC exam essays with hard word-count validation (India), references in ABNT/APA/MLA/SNI, thesis topic → outline → chapters following the country's academic standard • Spreadsheets: messy pasted text (WhatsApp lists, web tables) → clean table, spreadsheet analysis with summary stats, pivot and chart, ready-made template sheets (budget, ledger, timetable) • Photo & audio (vision + speech models): photo of a table → spreadsheet, receipt → expense report, photo of handwritten notes or a blackboard → structured notes, photo of a math or science problem → step-by-step study guide, handwritten essay → transcription + evaluation, audio recording → verbatim transcript + notes • Research: deep research that searches the web and cites sources • Tools: PDF merge/split/compress, format converter, e-commerce listing compliance check, mind map WHAT MAKES THE ARCHITECTURE INTERESTING • The user brings nothing. Most "AI agent" apps make the user supply a model key, an environment or a subscription. Pronto is a centralized service: the phone runs a thin Flutter client and one server does the LLM calls, vision for photo skills, speech-to-text for audio, and LibreOffice rendering. That is what makes it free and usable for this audience on a 2 GB phone. • Runs on one small server. No sandbox, no firewall rules for user code, no autoscaling story. The chat model can do exactly three things: recommend a registered skill, refine a previous result, or regenerate one. It cannot execute code, browse freely, or produce a deliverable outside a registered card, so the CPU, memory and tokens each job may use are known in advance. The whole product runs on a single mid-size VPS today, and it is easy to operate alone. • Billing that never needs a support inbox. Every task holds an estimate, settles to real usage and refunds the...