Truth-Guard

shipping it—TruthGuard-X is officially live and out in the wild.

Honestly, building this has been such a cool experience, even with the late-night grinds. Sorry I went a bit MIA lately—school submissions and deadlines have been absolutely relentless, so I had to pivot my focus for a minute there. But we’re back, the repo is fully updated, and you can finally check out the live version here: truthguard-one.vercel.app

The website is up and running smoothly, and while it’s in a good spot, I’m definitely going to keep an eye on things and push updates whenever a bug pops up (still looking at you, chat scroll logic). It’s been a massive journey from the initial “vibe coding” phase to a fully deployed platform with multi-model consensus and all those “Google effect” interactive characters.

If you’ve got a second, please drop a vote or a review I’d love to hear what you guys think of the final execution. It’s been a wild ride, and I’m hyped to finally have it in the hands of actual users. Catch you in the next project!

ok, honestly, TruthGuard is finally feeling like a finished product. I’ve spent so much time grinding on the backend logic that actually seeing the UI come together hits different. I got dark mode fully integrated and it looks so much cleaner now, plus the animations I added really bring the whole site to life. I also managed to get multi-language support in there we’ve got Hindi, Punjabi, Marathi, Bengali, and Tamil now, which was a huge goal to make it more accessible.

The login page for agent access is officially live too, and I’m lowkey obsessed with those little cartoon characters I added. I gave them that “Google effect” where their eyes actually follow your cursor around the screen it’s such a small detail but it makes the whole experience feel way more interactive and high-end.

The AI chat is still going strong with that multi-model consensus logic, even though I’m still squashing that annoying scroll bug and refining the response UI to make it look more aesthetic. I’ve also got the skeleton for the premium upgrade ready to go, just need to hook up the payment gateway later. But for now, the website core is basically done. The next move is finally getting this thing deployed and out into the world. It’s been a massive journey to get here, and I can’t wait to see people actually using it.

ok this will be a long one cuz quite a lot has been done, and honestly, the vision is finally coming to life. I just clocked a solid 8-hour session and TruthGuard is starting to feel like a real product now. Most of the website’s core functionality is basically wrapped up, which is a massive weight off my shoulders. The AI chat is officially live and working, and I even managed to get the skeleton for the “upgrade to premium” flow in there. The payment gateway side of things is still missing for now, but that’s the next big thing on the to-do list.

The AI chat is lowkey the star of the show. I didn’t just want a basic wrapper, so I went deep on the technical side using Next.js 15 and TypeScript. The logic is actually pretty cracked it’s hitting the OpenRouter API to run a multi-model consensus. Instead of just trusting one LLM, it’s basically a “Judge” system where it pings Claude 3.5, GPT-4o, and Llama 3 to cross-reference facts before it gives you an answer. It’s definitely making the misinformation detection side of the project feel way more robust and less like it’s just guessing. Getting that “Judge” logic to flow smoothly without hitting massive latency was a bit of a grind, but seeing it actually work in real-time is such a vibe. It’s crazy how much more professional the whole project looks now that the core features aren’t just placeholders anymore. Ready to keep this momentum going and get that payment side sorted next.

BUGLOG:
the chat isn’t perfect yet. It’s got a couple of bugs I need to squash, the biggest one being that I literally can’t scroll the chat right now once the conversation gets too long, it just cuts off. I’m also planning a major redesign for the response generation UI. I want it to look way more aesthetically pleasing while the AI is “thinking” and streaming the text in, instead of just having it pop up.

worked on the deepfake detection section and built out the overall flow. added the media upload interface, preview panel, and detection status UI so the feature now has a proper structure. also aligned everything with the CORE COMMAND dashboard to keep the intelligence terminal vibe consistent.

the USB upload is still a skeleton for now since this whole feature is experimental, but the base layout and interactions are in place and working. added some subtle animations and cleaned up spacing so it fits naturally into the workflow.

also started planning the main AI chatbot which is underway now. once that’s in, most of the core features will be in place and the project is slowly starting to feel like it’s coming together toward the final stretch.

the text snippet and url part is actually pulling real scores now but the screenshot upload was the real final boss. i kept getting these weird buffer errors when trying to pass the image data to the vision model and for a minute the whole dashboard would just flicker or go blank whenever i hit launch. i had to use my last two brain cells to optimize the state management so it wouldn’t re-render the entire page every time i uploaded a file… honestly a massive L for my sanity but a huge W for the website.

it’s lowkey satisfying to see a viral whatsapp headline actually get tagged as “suspicious” with a 70% score or whatever. still keeping it local for a bit while i clean up the error handling because i dont want the site to crash the second i push it but yeah the dashboard is finally alive. brain is buns rn, peace out.

spent about 4 hours building the CORE COMMAND dashboard today. started simple but ended up spending way too much time getting the cyber terminal vibe right. the grid background and spacing kept looking off so I kept tweaking opacity and layout until it finally felt clean.

also implemented the text snippet / url / screenshot tabs but ran into a weird animation bug where the active state kept jumping. fixed it after adjusting the state logic and smoothing out the transitions.

added the scan input box, core synchronized status, and launch intelligence scan button. mostly small UI work but the dashboard finally looks like a proper intelligence terminal now. next step is wiring it to the AI scan.

spent the last 2 hours just on the scroll reveal for the home page. i honestly thought it’d be a 20-min job but getting the sections to fade in perfectly without that weird stuttering was actual pain.

i’m using framer motion for the entrance animations but the viewport triggers were being so toxic like half the time it’d trigger too early or just stay blank. finally got the “staggered” look working so the cards pop in one by one and it looks way more “pro” now, but my eyes are legit burning. the landing page actually has some life in it now lol.

6 hours later and i’m finally done with the FIR assist stuff. honestly, this was such a bait i thought a 5 step stepper would be easy but keeping the data in sync from “Consent” all the way to “Location” was actual pain.

What I actually got done:
The logic for the whole flow is finally working. You go from the privacy check to the profile, and then the location part actually works. Seeing “Connaught Place PS” and “Lajpat Nagar” pop up with the right distances for Delhi was a massive vibe though. I also mapped out all the major helplines like 1091 and 1930 and linked them to the e-FIR portals so actually feels legit. Added a PDF download button at the end too just for that extra polish.

The struggle:
Spent like 2 hours just fighting with the form validation. If you skipped one field, the whole stepper would just glitch out and send you back to the start. Also, the CSS for the helpline cards was being so toxic on mobile it kept clipping through the background. I had to use way too much brainpower on coordinate math for the station distances, but we’re finally functional and it looks clean.

Finally moved past the landing page and mapped out the main sections: Report, Dashboard, Deepfake, AI Chat, and the Upgrade shop. Keeping the state from breaking while switching tabs was a whole mood.

But fr, the custom circle cursor was the absolute nightmare. I spent like 2 hours just trying to get the circle to stop lagging behind the actual mouse. It was so buggy.

Progress:
Tabs are live (Report, Dashboard, Deepfake, AI Chat, Upgrade).
Sticky circle cursor is (mostly) behaving now.

Tbh, I had to pull the plug on my old repo. I spent about 3 hours messing around with a fork of TruthGuard-2.0, but the more I looked at the code, the more I realized it was just… messy. The Git history was a disaster.

So, I’ve officially moved everything over to this fresh Next.js 15 setup. It felt weird “losing” those hours, but honestly, the architecture is 10x better now.
PS: I’m holding off on the big GitHub pushes for a bit—I want to make sure the core functionality is actually 100% solid before I start cluttering the commit history with a bunch of half-broken attempts. Will push the main update once the first “truth report” actually renders right.

What actually happened in these 5ish hours:

The Big Reset: Nuked the old repo and initialized this one. RIP old commits, you won’t be missed.

Next.js 15 + Tailwind v4: Got the basic scaffold up. Why is configuring environment variables always the hardest part? I spent 45 minutes just realizing I forgot the NEXT_PUBLIC_ prefix. 🤡

The AI “Brain”: Started wiring up the OpenRouter logic. I’m trying to build a “Judge” prompt that can actually spot the difference between a real headline and a viral WhatsApp forward that’s been “forwarded many times.”

UI Vibes: Got a clean dark-mode input box working. It’s simple, but it’s fast.

The Struggle: I spent way too long fighting with Framer Motion layout transitions. I just wanted the search bar to slide up smoothly when you hit enter, but it kept teleporting to the top of the screen. Finally fixed it, but my brain is fried.

Next up: Getting the credibility meter to actually animate based on the API response. 🚀

PS: I’m holding off on the big GitHub pushes for a bit—I want to make sure the core functionality is actually 100% solid before I start cluttering the commit history with a bunch of half-broken attempts. Will push the main update once the first “truth report” actually renders right.

Kicked off the TruthGuard-Pro rewrite today. Spent most of this session initializing the Next.js 15 environment and configuring the TypeScript types for our multi-model AI responses. I also managed to scaffold the basic dashboard UI with Tailwind CSS and got the environment variables synced for the OpenRouter API. It’s mostly just the ‘skeleton’ right now, but the project architecture is finally solid.