Before vs After
I want to let users simulate seeing their tweet in the X feed before posting. This is what i came up with.
I think it gives an added confidence before posting.
1
8
126
The Daily Ship - Tweet Preview ✅
Built a feed simulator so you can see how your tweet looks before posting it on X.
Small feature, big difference in confidence before hitting post.
2
94
New feature: Pinned Posts.
- Save drafts
- Edit them over time
- Shows the prompt used for each generation
- Preview how they'll look on X
- Mark them as posted
- Favourite posts to feed back into the AI.
Don't lose tweet ideas you could use later.
1
2
143
My stats for March are in.
My goal was to reach 100 followers & 350K impressions.
Started March with:
→ 57 followers
→ 188K impressions
→ 2.5% engagement
Ended March with:
→ 124 followers (+67 )
→ 445K impressions (+257K )
→ 2.8% engagement
The grind is working.
1
2
8
204
Small UI changes make a big difference.
Added quickfire questions so you can set your voice in seconds.
If you want more customization, there's text fields that act as your own personalisation prompt.
I want each users experience to be fully tailored to them, and give them the tools to achieve that.
1
135
Built XRA in a weekend. Posted it. Forgot about it.
68 people still installed it. 8 use it every week.
Once The Daily Ship launches.
I'm coming back to build XRA v2.
Two tools that work great together.
One for post ideas, the other for reply ideas.
Both built for devs on X.
1
4
120
3 months into X, just hit 100 followers.
Small number to some, but means a lot to me.
Thanks for being here😎
2
6
103
[5/5] Per generation overrides.
Your user's profile should stay constant.
But each generation might need a different tone, format, or length.
Profile = who they are.
Per generation settings = what they want right now.
This way one user can generate a sarcastic hot take and a professional build log without touching their profile.
1
58
[3/5] Anti-repetition.
AI has no memory between generations.
Without context, it repeats itself.
So I feed the user's recent outputs back into each prompt with instructions to:
• use different opening words
• vary the sentence structure
• contrast the tone and angle
1
33
[2/5] Compiled summaries.
Don't dump raw user data into prompts.
Instead, use AI to compile detailed inputs into something structured.
In the prompt, the raw data reinforces the summary, but the summary is the source of truth.
Here's what that looks like in practice.
1
40
The best way to humanize AI outputs?
Give the AI a better identity to work from.
Here's what I found actually matters for making AI sound like you.
(thread)
1
1
2
190
Personalisation separates usable AI outputs from slop. I'm focusing on this with The Daily Ship.
- Quick fire questions
- Text box for personal do's and don'ts
- Pasted voice samples (real tweets the AI studies)
This gets compiled into a voice profile that feeds into every prompt.
Here's how it's looking so far👇
1
2
105
3 months of posting on X.
I’ve found the hardest part is posting valuable content every single day.
So I started working on something to fix that for myself:
An app that turns your git commits, trending topics and your personalised profile into daily content ideas.
Built specifically for build-in-public devs.
Anyone else struggling with consistency?
How are you handling it?
1
1
1
173
“Just post consistently” is terrible advice if you don’t know what to post.
I’m building The Daily Ship.
It turns your profile, commits, and trends into actual content ideas.
Here’s the current landing page.
1
1
146
Here's a look at my progress on The Daily Ship so far.
Personalisation page that auto compiles a voice summary based on your inputs.
Single post generation that creates ideas based on your profile and customization.
Also working on a quick idea generator.
5 ideas you can expand into full tweets with one tap.
I'll have a basic MVP out soon for anyone that wants to give it a try.
1
123
What I’m doing differently building The Daily Ship
- Treat prompts like APIs → predictable outputs.
- Use examples → guide the model for better results.
- Inject user context → more personalised outputs.
- Design prompts like mini-programs → more reliable systems.
Small prompt changes can dramatically improve AI outputs.
1
87
4⃣ Your prompt is actually a mini-program
The more AI apps I build, the more I realize prompts aren’t just instructions. They’re small programs.
A good prompt includes:
• context
• rules / constraints
• examples
• structured output
1
36
3⃣ Personalisation changes everything
When I built XRA, the biggest problem was generic replies, they sounded like AI.
Detailed user context helps with this.
Things like:
• writing style
• interests
• past tweets
This made the replies feel more authentic.
1
45
1⃣ Treat prompts like APIs
Prompts shouldn’t feel like conversations.
When I built my support ticket summariser.
Vague prompts gave inconsistent outputs.
Treating the prompt like an API fixed it:
• define the task
• define the format
• enforce structure
1
3
136
AI apps are one of the most interesting things to build right now.
I've built:
• a support ticket summariser/translator
• a coding challenge generator
• an X reply assistant extension
Here are 4 prompt engineering lessons I've learned.
(Thread)
3
1
4
230
