Talaria

Inspiration

We’ve spent countless hours finding and applying to opportunities, and they completely changed our lives. One application helped Imen earn a €24,000 scholarship to engineering school, have her work featured in the Paris Metro, and opened doors that still create new opportunities years later.

We were already using ChatGPT to help us find and apply to opportunities. We realized we could automate much of that process and make every application smarter and more personalized. That’s what inspired us to build Talaria. We want to make life changing opportunities accessible to everyone.

What it does

Talaria is an AI agent that automates the process of finding and applying to life changing opportunities. Whether you’re looking for grants, fellowships, accelerators, incubators, scholarships, hackathons, or something else, Talaria helps you discover the opportunities that best match your profile and your goals. Every week, it recommends up to 10 personalized opportunities designed to help you get where you want to go.

To get started, Talaria builds a profile from your CV, LinkedIn, pitch deck, and anything else you want to share. The more you use it, the smarter it gets. It learns more about your experience, remembers information from previous applications, and adapts to your writing style over time.

When you’re ready to apply, simply upload the application in any format. Talaria generates personalized answers, asks for any missing information, and helps you improve every response before you submit it yourself.

How we built it

Codex became our engineering teammate. Every time we found a bug, it followed the same workflow. It opened the live product, traced the issue, wrote the fix, deployed it, then tested everything again in production from a real user’s perspective. Every coding session was powered by GPT 5.6.

Finding opportunities is not just a single prompt. It is a complete pipeline. We start with trusted official sources, then use GPT 5.4 mini with web search to expand the search when needed. Every opportunity is checked to make sure it is still active, that you are eligible, and that the application link works. If something is broken, Talaria fixes it and verifies everything again before moving on.

Only verified opportunities reach the final ranking step, where Talaria decides how well they match your profile and explains why they are a good fit. It can reject an opportunity if it is not relevant, but it can never bypass the trust and verification checks.

The application copilot also keeps every version of your work, so if you regenerate an answer, it never overwrites something you wrote yourself.

Challenges we ran into

One of our biggest challenges was finding reliable application links. Many opportunities had broken, outdated, or incorrect links. We couldn’t just trust the first result, so we built a verification system that checks every link, repairs broken ones when possible, and makes sure it points to the right application before showing it to the user.

We also realized that opportunities change all the time. Pages move, deadlines expire, and many websites don’t clearly show when information is outdated. That’s why we built a repair and verification pipeline instead of relying on a single search.

Another challenge was making the application process reliable. Sometimes a CV or document was only partially processed. Instead of making users start over, we made every step safe to retry without losing any progress.

We also ran into a limitation with websites that require an account, like LinkedIn or Notion. Since the AI can’t always access that information, we added a simple prompt that users can paste into ChatGPT to extract the missing information and import it back into Talaria in seconds.

Accomplishments that we're proud of

One thing we learned is that verification really matters. It would have been much easier to build a search engine that returns plausible opportunities, but that’s not enough. Every opportunity is verified before it reaches the user.

We also built a complete audit trail. We can replay every recommendation step by step and see exactly why an opportunity was accepted or rejected. That made debugging much faster and even turned one production bug into a one line fix and a regression test.

From day one, we decided Talaria would never submit applications for users. The goal is to help people write stronger applications while keeping them in control of every submission.

We also tested Talaria with 14 early users. Every participant completed onboarding, and 100 percent finished setting up their profile. People explored dozens of recommended opportunities, opened 42 percent of opportunity details, clicked through to 25 official websites, and many saved opportunities they wanted to apply for later. The feedback was very positive and showed us that users were finding opportunities they genuinely cared about.

What we learned

We learned that an observable pipeline beats a clever prompt. Every real problem stayed invisible until we could see exactly which step dropped an opportunity. Once we could, most bugs became simple to fix.

We also learned that verification is the product. Finding opportunities is the easy part. Making sure they’re real, up to date, trustworthy, and actually a good fit is what really matters.

Another lesson was that people often try to be nice instead of telling you what’s actually wrong. Having Codex as a beta tester helped us catch issues users never mentioned. And building with continuous feedback instead of waiting until the end helped us improve the product much faster.

What's next for Talaria

Next, we want to expand our discovery engine with more sources, especially regional programs, university fellowships, and private communities where many of the best opportunities are shared.

We also want Talaria to learn from every application by tracking what gets accepted, what gets rejected, and what actually leads to life changing outcomes, so it can make even better recommendations over time.

We’ll continue testing the product with more users, collecting feedback, and improving every part of the experience through rapid iteration.

At the same time, we want to launch a waiting list to measure demand, learn who wants Talaria the most, and start building a community around it.