Tech • AI • Robotics • Game
Kimmy K3 finance tracker and GPT6 Astra Form app builds
Briefing
Kimmy K3 builds finance tracker
Kimmy K3 was used to create a browser-based personal finance tracker from a blank chat, with no starter code, local tooling or API setup. The result was a standalone single HTML file using Tailwind CSS and vanilla JavaScript, aimed at low-friction prototyping. Core functions included transaction entry, balance calculation and a mobile-friendly interface. The workflow highlighted how a free browser-based model can move from idea to usable software in staged prompts.
GPT6 Astra shapes Form
GPT6 Astra was used to build Form, a premium-styled workout tracker, through a five-prompt workflow. The app launched with a dark theme, lime green accents, a sidebar, and dedicated Log and Dashboard views, giving it product-level structure from the outset. Rather than a demo widget, the result resembled a cohesive fitness application with navigation, stateful logging and analytics. The build underlined how prompt sequencing is becoming a practical interface for full app scaffolding.
Finance app stays single-file
The finance tracker centered on a single-file architecture, packaging UI, styling and logic together for easy iteration and deployment. Initial features included a running balance header, transaction form, category selectors and a prefilled transaction list. Categories such as food, utilities, salary, entertainment and transport were surfaced with color-coded badges for fast scanning. That design choice favored accessibility and portability over framework complexity.
Form adapts by workout type
Form changed its logging inputs dynamically depending on whether the session was strength, cardio or flexibility. Strength workouts exposed sets, reps and weight, cardio switched to duration and distance, and flexibility reduced the flow to duration only. This adaptive behavior made the product notably closer to real training workflows than fixed-form trackers. It also demonstrated more reliable state and conditional UI generation from model-directed app building.
Analytics become live in both apps
Both projects moved beyond static layouts by wiring calculations directly to user actions. In the finance app, summary cards tracked total income, total expenses and net balance, with styling that changed when balances turned negative. In Form, dashboard cards calculated total sets, total reps, total weight in kilograms and workouts completed this week from actual logged sessions. The common thread was the shift from mock interface generation to functioning client-side analytics.
Instant updates anchor usability
The finance tracker refreshed the transaction list and running balance immediately after a new entry, avoiding page reloads and reinforcing its utility as a day-to-day tool. Testing with sample purchases like grocery spending showed the app could append records and recalculate totals in real time. Form similarly tied its dashboard to logged workout history so changes surfaced directly in stats and charts. These behaviors matter because responsiveness is often the dividing line between a prototype and a usable product.
History, PBs and streaks deepen Form
Later iterations of Form expanded beyond basic logging into retention-oriented fitness features. The app added personal-best detection, smarter history controls and streak tracking, layering motivation and review tools onto the base experience. Those additions are significant because they move the product closer to the engagement loops used by commercial workout software. The result was a more complete consumer-style application rather than a one-screen exercise logger.
Prompt staging drives app quality
Across both builds, the most notable pattern was the use of tightly scoped, sequential prompts instead of a single all-in-one request. Kimmy K3 reached a polished finance tracker in four stages, while GPT6 Astra assembled Form in five prompts, each adding a distinct functional layer. That incremental method improved structure, reduced ambiguity and made feature expansion more controllable. It points to an emerging best practice for AI coding workflows: build the shell first, then add logic, analytics and refinement in passes.
Videos covered
How to Build Your First App with Kimi K3 (FREE)
- •Beginner-focused setup
- •Single-file finance tracker foundation
- •Categories and live transaction updates
How to Build Your First App with GPT 6 Astra In 18 Minutes
- •From idea to product structure
- •Adaptive workout logging
- •Dashboard connected to real data