This is a submission for the Built with Google Gemini: Writing Challenge
Future Me — A Terminal That Lets You Talk to Yourself One Year From Now
What I Built with Google Gemini
Most productivity apps track what you did yesterday.
I wanted to build something that talks to who you’re becoming.
So I built Future Me — a retro-inspired terminal interface where you input your current goals, struggles, skills, and a message to yourself… and Google Gemini generates a response from “you,” one year in the future.
Not predictions.
Not generic motivation.
A structured, reflective projection of who you could become — based on who you are today.
The app asks for:
- Your current goals
- Your current struggles
- Your skillset
- A short message to your future self
Then Gemini generates:
- A letter from your future self
- Growth highlights you’ll achieve
- Habits that will hold you back if unchanged
- A practical 30-day action roadmap
I deliberately designed the UI as a retro terminal instead of a modern AI SaaS interface. It feels like you're connecting to a time machine rather than filling out a form.
The Role of Google Gemini
Google Gemini (using gemini-2.5-flash) powers the entire reflective engine.
It doesn’t just rephrase input. It:
- Detects patterns in goals and struggles
- Infers behavioral risks
- Projects growth trajectories
- Creates emotionally grounded but practical responses
- Maintains structure with clearly defined output sections
The key wasn’t just “call the API.”
It was prompt engineering for introspection, realism, and specificity.
I structured the system prompt so Gemini responds as:
“Your future self, exactly one year ahead, speaking honestly but constructively.”
That framing changed everything. The output became less motivational and more grounded.
The app is deployed on Cloud Run, containerized via Docker, and scales to zero when idle.
Demo
You can try it here:
The application is deployed on Vercel and built as a lightweight React + Tailwind SPA.
What I Learned
1. AI feels more powerful when it’s constrained
The first version of the prompt was too open-ended. The output was poetic but vague.
Once I structured the output into:
- Letter
- Growth Highlights
- Habits to Fix
- 30-Day Plan
The results became dramatically more actionable.
Constraint improved creativity.
2. Tone matters more than temperature
Small changes in instruction tone had massive effects.
“Be motivating” produced generic output.
“Be honest, reflective, and slightly uncomfortable when necessary” produced depth.
Gemini responds strongly to psychological framing.
3. Retro design changes how users perceive AI
The first UI looked like every other gradient AI tool.
Switching to a monochrome terminal interface:
- Removed the “startup template” feel
- Increased immersion
- Made the experience feel intentional
Design influences how seriously users take AI output.
4. Deploying to Cloud Run is underrated
The deployment was surprisingly simple:
- Dockerized Vite build
- Exposed port 8080
- Deployed via
gcloud run deploy --source .
Cloud Run’s scale-to-zero makes side projects economically sustainable. That alone changes how willing you are to experiment.
Google Gemini Feedback
What Worked Well
1. Structured Output Consistency
Gemini handled clearly sectioned prompts extremely well. It respected formatting boundaries and maintained narrative voice across sections.
2. Psychological Inference
Even with short inputs, Gemini identified behavioral patterns surprisingly accurately — especially around procrastination, fear of failure, and focus issues.
3. Speed
Using gemini-1.5-flash, response times were fast enough that the experience felt real-time.
Where I Felt Friction
1. Over-Optimism Bias
Without careful prompt tuning, responses defaulted to “you will succeed.” It required deliberate instruction to introduce realism and constructive critique.
2. Client-Side API Exposure
For simplicity, I called the Gemini API directly from the frontend. This exposes the API key in the client. In a production environment, I would proxy requests through a backend service.
3. Subtle Output Drift
Even with the same system prompt, tone varied slightly between sessions. It’s a reminder that prompt engineering is ongoing maintenance, not a one-time setup.
Looking Forward
Future Me started as a challenge project.
But building it changed how I think about AI tools.
The most powerful use of generative AI isn’t replacing work.
It’s accelerating reflection.
We measure performance constantly.
We rarely measure direction.
This project attempts to do that.
If Gemini continues improving in contextual reasoning and structured generation, tools like this could evolve from novelty to personal decision-support systems.
For now, it’s a time terminal.
And sometimes, that’s enough.
Top comments (0)