Bloomshot: Water Rescue

Inspiration

I wanted to turn my simple water-routing prototype into a polished mobile puzzle game with smooth digging and playful, readable graphics. Classic dig-and-route puzzle games inspired the immediate touch interaction, but I wanted Bloomshot to have its own garden-rescue identity and a mole that changes the terrain during play.

What it does

Bloomshot is a portrait mobile puzzle vertical slice. The player drags through soil to carve a continuous channel, guides fresh water out of a reservoir, routes it around indestructible stone, and restores a dormant flower. A small mole independently digs in random directions inside a protected area. Its tunnel is temporary: older sections gradually refill after about ten seconds, creating a different route pattern each run without destroying the authored solution.

How I built it

I rebuilt the project in Unity 6 with C#. The game separates an engine-independent deterministic simulation from Unity presentation code. The core uses a fixed-step grid for terrain, digging, water movement, target absorption, scoring, and level validation. Unity's Universal Render Pipeline, Input System, uGUI, particles, custom artwork, and a board-composite shader provide the portrait presentation and smooth the grid simulation into continuous-looking tunnels and water.

I used OpenAI Codex with GPT-5.6 throughout the rebuild for genre research, architecture planning, implementation, automated tests, visual iteration, playtest diagnosis, and repeatable Windows and Android build automation. GPT-5.6 is part of the development workflow, not a runtime dependency; the shipped game runs locally and sends no gameplay data to an AI service.

Challenges I ran into

The hardest part was making grid-based water and digging look smooth while keeping the puzzle deterministic and solvable. The mole also needed several iterations: it was initially too large and static, then its random paths could pause or interfere with the level. I constrained its roaming area, tested many seeds, reduced its visual scale, and made the trail refill gradually.

Accomplishments that I'm proud of

I'm proudest of the mole system: it digs in random directions and leaves a temporary trail that refills after about ten seconds. The current build is a complete one-level vertical slice with touch digging, flowing water, custom garden artwork, animation, particles, a bloom goal, Windows and Android builds, and eight automated tests covering fluid mass, rock behaviour, target absorption, solvability, timing independence, and safe random mole paths.

What I learned

A deterministic grid can still feel organic when the renderer hides cell boundaries and feedback is layered carefully. Random variation is most useful when it is constrained by authored safety rules. Repeated visual playtests were just as important as simulation tests.

What's next

Next I would add a full level progression, new water and terrain mechanics, onboarding, sound and music, accessibility options, save progression, localisation, and release-quality device testing.

Built With

• c#
• gpt-5.6
• openai-codex
• unity-6
• unity-input-system
• universal-render-pipeline