A few months ago I shipped https://t.co/ukbx3phxLp, a parent-child screen-time earn with rewards tracker I built because I wanted to solve a problem in our house. That was my first AI build project, and it taught me what working with Claude as a coding partner is actually like. I've been really hooked since.
For another project, I wanted to keep the personal-utility bar high, but go further on AI being the central feature, not just the tool to build with.
I built https://t.co/EekoitdLcG. Our family has two older vehicles we love and want to keep running. I'm not someone who meticulous service records, I'm reactive instead of proactive about maintenance, and every time I'm at the service center I get surprised by what they say I need. I wanted something that would tell me what's actually due, what's coming up, and what's based on the manufacturer's schedule versus a shop's upsell.
Five different AI integrations make this work, and each one solves a problem AI is really good at:
Claude reads receipts. I can upload a photo or email it to the [email protected] forwarding address, and Claude extracts the vendor, date, mileage, services, line items, and cost. It reads the receipt like I would.
Claude matches receipts to vehicles. We have multiple cars, and the emailed receipt gets matched based on what's on it (our VINs and year/make/model) cross-referenced with the cars I’ve onboarded.
Claude decodes VINs from photos. I can enter this manually, but thought it would be a cool 1 time set up feature to point my phone at the VIN sticker on the door jamb, where Claude reads the 17-character string. Then NHTSA's free public API decodes it into year, make, model, trim.
Claude categorizes services. Service shop receipts use variations of language: "Cooling System Exchange," "Coolant Flush," "Antifreeze Service" all mean the same thing. Claude classifies each service into a canonical maintenance category so the engine actually understands what's been done.
Claude knows manufacturer schedules. With the vehicles I’ve added, Claude returns the OEM-recommended intervals for that specific make, model, and year: synthetic oil at 10,000 mi for a Wrangler, 100,000+ mile coolant intervals where the manufacturer specifies them, brake fluid at the right cadence per spec.
The result is the maintenance road ahead I needed to know both what services I’ve done and what our vehicle manufacturer actually recommends, instead of generic 5,000 mile oil change reminders that don't match any modern car.
The stack for https://t.co/EekoitdLcG:
#React #Vite frontend
#Supabase for auth and data
#Vercel for serverless functions
#Postmark for inbound email forwarding
NHTSA vPIC for VIN decoding
#Claude (Haiku) doing the cognitive heavy lifting across five request calls
The site runs at near-zero cost per user because each AI call is small and cached where possible.
Try it out if you’ve been needing something similar!
Just tested this and I’m impressed. Integrates cleanly with our existing apps and enables rapid, realistic prototyping on top. It hits the sweet spot Lovable is great for as well, but in a way that feels actionable quickly. Going to be using this a lot!
https://t.co/IAcG5JDWeX
We are building Reforge Build for product teams.
Try it here: https://t.co/Padse2fpTA
Like + Repost the original video tweet, then DM me. I’ll send you a free month of premium and some additional AI credits.
➤ Prototypes that look like your product
➤ AI that remembers your customers, feature set, and strategy
➤ Plan mode that structures ideas before generating
➤ Generate multiple variants, not just one output
➤ Built-in collaboration with sharing, commenting, and remixing
➤ Validate with customers w/ AI Interviews and Surveys (coming soon)
The difference?
AI App Builders: Start From Scratch → Build An App → Deploy + Host
Reforge Build: Start From Your Product → Explore Variants → Collaborate → Validate
If you’re a product team not building your first app, stop using tools that assume you are.
Other tools are built for starting from scratch. We are building for the reality of product teams.
@engineering_bae@adambain — who opened the product door for me which was a great fit. My prior dev/design roles were good match for my curiosity but not really my skillset. Forever thankful!
Wrote little eulogy to Flash, before it officially dies at midnight in a few days. Many of us who grew up designing in the '90s wouldn't be where we are today without Flash. I, for one, will always be thankful. https://t.co/kgoX3HZmLj