raphGPT Feature Drop
24 June 2025
Recently I have been getting feedback from users saying things like
- Timezones are wrong in other countries
 - Taking too long to respond
 - Errors when sending location
 - Files like PDF, DOCX not supported
 - Asking for directions is like asking a wall. It just sends a link to Google Maps.
 - Would be great to have ability to schedule messages
 - Incorrect location analysis, given a link to another foreign place rather than the destination.
 - Inconvenient for international users, have to configure language. Would be great if it asked the first time.
 - Sometimes doesn’t know what to do when the same message is sent multiple times in a row
 - Sending more than 1 photo at a time causes it to reply for x amount of photos there are
 - Need to make it easier to find bus stops by name
 
I took some time to think about it and took all the feedback in.

With that said, I am excited to announce a feature drop today.
In this feature drop, I tackled each piece of feedback head-on.
Here’s a breakdown in order of implementation:
- Lean Summarization & Google Lens
- Switched to a lighter summarization model for speed
 - Added Google Lens integration for better image queries
 
 - SuperJSON Serializer/Deserializer
- Improved data handling for richer context
 
 - File Agent & Improved File Handling
- Full support for PDF, DOCX, and other formats
 - Media-group support so multiple files/photos are handled in one go
 
 - Directions Tool & Timezone Fixes
- Built a native directions tool instead of linking out
 - Continued execution after timezone changes and fixed incorrect offsets
 
 - Message Scheduling & Context History
- You can now schedule messages easily
 - I store Codex outputs in history, so context is never lost
 
 - Language Handling First-Run Prompt
- I’ll ask your preferred language on the first run, then remember it globally
 
 - Error Handling & Robust Tool Calls
- Proper language error handling with friendly prompts
 - Retries on tool-call failures to reduce friction
 
 - Bus Stop Matching
- Reduced threshold for matching bus stop names in the LTA agent
 
 - Cleanup & Chores
- LTA agent cleanup, interruption pauses, system message tweaks, and more
 
 

What’s Next?
- Faster Response Times
 - Agent framework
 - Planning, then thinking process — Describe, Explain, Plan, Select
 - Advanced Routing: More transit options (biking, walking) and departure reminders.
 - Location Accuracy: Partnering with mapping providers for pinpoint precision.
 - Interactive Tutorials: Quick in-chat demos for new features.