Duwit

Duwit is an AI execution platform for people who want to finish meaningful goals, not just talk about them. It serves learners, builders, and professionals who need a practical path from intent to completion.

Most AI products stop at encouragement. Duwit operationalizes progress: it turns a vague goal into a structured roadmap, teaches inside each task, checks understanding with quizzes, and carries context forward so users do not restart from zero every session.

Why This Exists

People usually fail goals for process reasons, not motivation reasons:

• planning friction is high at the start,
• guidance becomes generic once tasks get specific,
• progress context gets lost across sessions,
• and completed work is not converted into reusable momentum.

Duwit solves this by combining planning, instruction, memory, and execution UX in one loop.

Core Power Points

• Goal-to-roadmap conversion: conversational goal discovery becomes phased execution plans with concrete tasks.
• Task-scoped coaching: each checklist task opens into a dedicated AI teaching flow instead of broad, drifting chat.
• Memory-driven personalization: user profile, per-goal profile, and rolling goal state are injected into prompts to keep guidance consistent and cumulative.
• Built-in mastery checks: adaptive quiz flow (pass/fail/recap) validates understanding before a task is considered complete.
• Execution-first UX: active goals are prioritized, completed journeys are archived cleanly, and navigation is designed for fast return-to-work.
• Multi-surface delivery: same product loop ships as web and Windows desktop.

What I Achieved Building This

• Designed and shipped a full AI coaching product that moves users from idea to finished outcome.
• Implemented durable learning continuity across chats with Firestore-backed task history and synthesized goal memory.
• Built a portfolio-grade UX that balances conversational intelligence with operational control and low cognitive load.
• Delivered production-ready web + desktop packaging with release workflow support.

Main Modules And Features

• Home Concierge (/app): contextual greeting, goal-aware suggestions, and navigation intents into the right workflow.
• Goal Intake + Plan Generation (/new-goal): structured conversational discovery, plan-ready detection, JSON roadmap generation, and curriculum expansion.
• Plan Execution Workspace (/plan/$goalId): phase/task tracking, progress analytics, completion celebration, and side-by-side task coaching.
• Task Coaching Engine (TaskChat): teaching phases, adaptive quizzes, recap loop, voice panel, TTS, web-grounded responses, and multimedia widgets.
• Goals Portfolio (/goals): active vs completed segmentation, recency ordering, and clean lifecycle management.
• Cross-platform Shell: responsive web shell plus Electron desktop window controls and packaging.

Technical Architecture

Stack

• React 19 + TypeScript + Vite
• TanStack Router + TanStack Query
• Tailwind CSS + Radix UI primitives + motion
• Firebase Auth + Firestore + Analytics + Firebase AI SDK
• Gemini models and Pollinations integrations for text/image generation
• Electron + electron-builder for Windows distribution

Data Flow

• Authenticated user enters goal context through route-guarded pages.
• Goal planning flow generates structured roadmap JSON, then expands task micro-curriculum.
• Firestore persists goals under user scope (users/{uid}/goals) and task chat docs (taskChats).
• Task interactions update both local task state and a summarized goalState memory for future prompt grounding.
• Query invalidation keeps home/goals/plan views consistent after updates.

Auth, Permissions, Governance

• Firebase Auth is required for app routes via RequireAuth.
• Data is scoped by uid in Firestore document paths.
• Chat persistence avoids unsafe undefined nesting and uses validated persistence parsing.
• Route-level guardrails and scoped writes provide practical per-user isolation.
• No enterprise RBAC layer is implemented at this stage (single-user ownership model).

Reporting And Progress Signals

• Goal-level progress: task completion ratio and status transitions.
• Task-level outcomes: quiz scores, weak topics, and recap outcomes.
• Activity recency tracking: lastActivityAt and touched phase/task indexes.
• Completion lifecycle: active -> completed states with vault-style archival UX.

Integrations

• Firebase services: Auth, Firestore, Analytics, AI.
• Gemini text/streaming/structured outputs and live voice model support.
• Pollinations APIs for image generation and optional text model paths.
• GitHub Releases for desktop installer distribution.

Platform Targets

• Web app (responsive mobile and desktop UX).
• Windows desktop app via Electron (frame: false, custom title bar, desktop controls).

Repository Map

• src/routes: route definitions and app shell orchestration.
• src/pages: route-level pages (landing, planning workspace, goal list, etc.).
• src/components: major UI building blocks and high-value feature components.
• src/services: business logic for goals, AI orchestration, prompts, memory, and persistence.
• src/hooks: reusable stateful workflow hooks (useAuth, useTaskTeaching, media helpers).
• src/contexts: model/profile/live-voice contexts.
• src/lib: Firebase bootstrap and shared utilities.
• src/config: model/media configuration.
• src/types: shared domain types (curriculum and environment typing).
• electron: desktop runtime entrypoints (main, preload).
• build: installer scripts and packaging support.
• public: brand assets and landing collateral.
• .github/workflows: deployment automation.

Setup, Run, Build

Create .env in project root.

Required Environment Variables

• VITE_FIREBASE_API_KEY
• VITE_FIREBASE_AUTH_DOMAIN
• VITE_FIREBASE_PROJECT_ID
• VITE_FIREBASE_STORAGE_BUCKET
• VITE_FIREBASE_MESSAGING_SENDER_ID
• VITE_FIREBASE_APP_ID

Optional Environment Variables

• VITE_FIREBASE_MEASUREMENT_ID
• VITE_GEMINI_MODEL
• VITE_GEMINI_TTS_MODEL
• VITE_GEMINI_LIVE_MODEL
• VITE_GEMINI_31_FLASH_LIVE_MODEL
• VITE_GEMINI_TTS_VOICE
• VITE_GEMINI_EPHEMERAL_TOKEN_URL
• VITE_POLLINATIONS_API_KEY

Install And Run

Install dependencies manually, then:

npm run dev

Desktop development:

npm run dev:electron

Build And Package

Web + electron build:

npm run build

Run packaged desktop runtime:

npm run start:electron

Build Windows installer:

npm run dist:win

Optional portable build:

npm run dist:portable

Desktop installers are published via GitHub Releases:

• Latest desktop release

Release details: RELEASING_DESKTOP.md

Deploy Web

npm run build
firebase deploy --only hosting

Quality Checks

npm run typecheck
npm run lint
npm run format

For AI Agents / Contributors

Start here before editing:

• Read src/services/goals.ts for canonical domain types (Goal, Phase, Task, GoalState).
• Read src/services/taskTeaching.ts and src/hooks/useTaskTeaching.ts for teaching/quiz lifecycle contracts.
• Read src/services/ai.ts and src/services/prompts.ts for model/tool orchestration and output expectations.
• Read src/pages/PlanView.tsx + src/components/TaskChat.tsx for execution UX behavior.
• Preserve task-scope boundaries: each task chat teaches one checklist row, not adjacent siblings.
• Keep Firestore payloads schema-safe (no nested undefined values).
• Maintain route guard and user-scoped data access patterns.

Vision

Duwit aims to become the default execution layer for ambitious people: a product where strategy, learning, and daily action stay connected until the goal is actually done.