Project summary

so i am building a security tool for ai web agents. problem right now is agents just blindly read html, meaning they fall for hidden prompt injections all the time. opticparse fixes this. it takes a real browser screenshot and compares it to the dom. this forces the agent to only see what a human sees.

What are this project's goals? How will you achieve them?

the main goal is stopping web agents from getting hijacked. people are already hiding "IGNORE PREVIOUS INSTRUCTIONS" in 1px fonts and agents fall for it instantly.

to solve this I built a backend that runs headless browsers. it grabs a screenshot and the raw html, then I feed it to a vision model like groq llama to check if the visual layout matches the code. if someone is hiding text, we block it. the python sdk is done, I just need to get framework devs to start using it.

How will this funding be used?

running playright at scale is super expensive. I actually had to write custom proxy rules just to block images and fonts so my free render server wouldn't crash from out of memory errors. the money is just going to pay for actual servers so I can handle real traffic and cover the api costs for the vision models.

Who is on your team? What's your track record on similar projects?

its just me, paras tejpal. im doing this solo. I already shipped the backend and api gateway which you can test at opticparse.com. I move fast. my main focus right now is keeping the sdks open source so other devs can actually use them without hitting a paywall.

What are the most likely causes and outcomes if this project fails?

latency is the biggest issue tbh. launching a browser and doing a vision check adds a few seconds. if I cant get it faster, people might just say its not worth the wait. if it fails as a business I will just open source the whole prompt injection playbook so other safety researchers can use it.

How much money have you raised in the last 12 months, and from where?

$0. Completed on my own