Archetype

Inspiration

As university students, we see and understand the stress fresh graduates face upon graduation: "What do I do next" or "Where do I start?" Therefore, we understood the importance of a tool that mentors graduates, clarifying the next steps to help them navigate their future with confidence. Because as most industry professionals say, “What you learn in your universities, is not what you do in work”.

What it does

The tool takes important data from the fresh graduate such as the graduate’s CV or background, interests, skills, willingness to pay for training, and time until employment, skills, interests. Using this information, it gives the user certain job paths and several roles under each path. With every decision, the tool shows the AI thinking behind it to remain as transparent as possible. There is also the option for the user to correct the AI’s understanding of the user’s interests and aspirations. After the user’s selection of the most appropriate path, the tool generates a weekly timeline of tasks to do, including videos, play-lists, training and courses to provide all the essential information about the selected job, with links to trustworthy sources based on the API’s and databases.

How we built it

Archetype is built as a React frontend deployed on Vercel and a FastAPI backend deployed on Railway. When a user uploads their CV, the backend extracts the text and passes it to Claude (Anthropic's API), which interprets their background and maps them to career fields and roles. Each role is grounded in real O*NET occupational data before the AI reasons over it, so the skills and tools it highlights are verified facts rather than generated guesses. From there, Claude sequences a personalised learning roadmap, and SerpApi pulls live job postings and real course results to populate the openings and resources sections. The entire pipeline follows a reasoning over retrieval design: the AI never invents verifiable information, it only interprets and explains data that came from an authoritative external source.

Challenges we ran into

The biggest challenge was keeping the AI honest. Early versions let the model generate salary figures, course links, and job listings directly, which meant it would confidently fabricate plausible-sounding but fake information. We had to redesign the entire pipeline so the AI only reasons over data retrieved from real sources, never inventing anything itself. On the data side, O*NET's skills survey does not cover every occupation, so newer roles like Data Scientist returned empty results and we had to build a backfill system that borrows skills from the nearest related occupation and flags that transparently. Deployment also threw surprises, with the frontend baking the localhost URL into the production bundle because Vite reads environment variables at build time rather than runtime. The design was its own challenge entirely. We went through multiple complete restarts before landing on something that felt genuinely easy to navigate, and a lot of the work was about resisting the urge to add more and instead stripping things back so the user never feels lost.

Accomplishments that we're proud of

When we started, Archetype was just an idea on paper. We had a vision for how it could work but turning that into a real, interconnected system felt genuinely difficult, and there were moments where it was not clear how all the pieces would fit together. Finding solutions one by one, watching the pipeline actually come to life, and then seeing it all work end to end was something we did not take for granted. Being able to submit exactly what we set out to build, not a stripped-down version of it, is what we are most proud of.

What we learned

We learned about how we can deploy a full stack application across two separate cloud platforms, Railway for the backend and Vercel for the frontend, and it taught us how small configuration details can break the build in ways that never show up locally. Integrating multiple APIs and datasets, each with their own formats, rate limits, forced us to think carefully about how to structure a system that stays maintainable as more pieces get added. The biggest conceptual lesson was around AI architecture: understanding the difference between what a language model should do versus what it should never be trusted to do. Designing the reasoning over retrieval split, where the AI interprets and explains but never fabricates verifiable facts, changed how we think about building with AI entirely.

What's next for Archetype

For our tool to operate for a larger audience, we are thinking of including additional features for future premium users. Also, adding a regenerate button or feedback box after the timeline would allow the user to further adjust or customize the planned timeline easily. Additionally, we are thinking of adding more datasets to make this work accurately for people across different regions, not just the current O*NET dataset which is US based.