Inkline

• 

  messy input

• 

  vs code open ai model api key

• 

• 

  typing

• 

  send

• 

  polishing

Inspiration

Ground it in something real and specific , not abstract . example : many people typing into chatgpt / cluade in their native language , with typos or vague phrasing , get worse answers than they should - the ai is doing extra work parsing bad input instead instead of helping . Grammerly fised writing : nothing fixes prompts the same way everywhere you type them

What it does

1. you type a prompt - any language , any grammer .
2. Hit Polish -> if detects language , translate if needes , restructures for clarity
3. see a redline off (removed /added ) before accepting - you stay in control
4. Accept or Discard then send into the chat .

How we built it

React Frontend , GPT - 5.6 for the optimize + chat calls ,your own word -level LCS different algorithm for the redline view (this is genuinely interesting technical detail worth 2 -3 sentences - most people just show a before /after , you show an inline diff) . Mention codex specifically here - what you used it for ( scaffolding , debugging , the diff algorithm ,etc) , since judges are scoring .

Challenges we ran into

Pick 1-2 real ones . Good candidates getting the diff algorithm to feel like an editor 's markup instead of a wall of red / green ; handling non latin scripts(tamil ) correctly in the diff view since word - splitting behaves differently ; keeping latency low enough that polish doesnt fell like a wait

Accomplishments that we're proud of

The redline / margin -not UI ,since its the one thing that isnt generic - most "Prompt Improver " tools jsut show plain before after text.

What's next for Inkline

Browser extension (inject Polish button into any site's textbox), mobile keyboard extension, desktop overlay — you already have this roadmap from earlier, just compress it to 3-4 lines.

Built With

• codex
• gpt-5.6
• javascript
• lucide-react
• node.js
• openai
• react
• tailwindcss
• vercel