Yup, easiest project of my life.
Step one? Interview ~100 customers to figure out what the hell to build.
Then it took just 3 engineers, 1 UX/UI, 6 months, and the complete loss of any concept of “weekend.” ☠️
We had to… you know, build our own RAG, test a couple of vector DBs, try some ML algorithms, and play with a few half-baked agentic frameworks.
I'm glad AI is such a non-noisy space — the choices are obvious and everything works exactly like it says in the outdated documentation from 110 releases ago. 🤡
Then came function calling, plus custom tools so the agent could actually do grant stuff.
Turns out “Write me a grant application” actually requires… consultant-level skills. Who knew? 🧠
Eventually we asked ourselves:
Where’s the data coming from? Your standard, everyday crawler? Pfft, adorable. 🤖
So we built a generic AI crawler agent that can extract grants from any government portal in any country.
And then — naturally — one to submit applications too. Just your typical AI wrapper work.
8 agents in total so far.
Naturally, we spun up 2 testbeds and brought in patent lawyers to make sure we weren’t completely out of our depth. ⚖️
By then, the tests were saying our AI was… let’s say… a little slow... 🐌
So we brought in dozens of grant experts to help fine-tune it.
But things were still too simple, so we thought:
What if we went BANANAS and wrapped more than one LLM??? 🍌
So we did:
Mini for quick wins, DeepSeek for casual heavy lifting, LLaMA because the name is fun.
A router and a few logic rules decide who gets to speak.
And of course, this doesn’t even mention the "old-school" tech struggles — our shiny new frontend JS framework to wrangle SSEs; dealing with legacy leftovers from our previous pivot, and the dev-ops required to glue together dozens of microservices. 🔧
This is the simple stuff I actually enjoy. Deep down, I’m just a simple monolithic country boy.
All of this…
For a GPT wrapper.
Exciting times. ✨
I used to build code wrappers.
But this one?
This one has opinions.
Join the new Granter here
https://lnkd.in/dmNwxSxs
This post was originally shared by on Linkedin.