UNMAPPED

Inspiration

The inspiration for UNMAPPED stems from the "invisibility" of the global South’s workforce. Millions of highly skilled individuals—from electronics repair technicians in Accra to mobile developers in Lahore—operate within informal economies where their expertise is never formally documented or recognized. We wanted to build a bridge that translates this real-world hustle into a language that the global labor market understands.

We were moved by the realization that while AI is often seen as a threat to jobs, it can also be the ultimate tool for inclusion. By using AI to map informal experience to international standards like the ESCO taxonomy, we can give every worker a "portable professional identity" that doesn't rely on a traditional university degree.

What it does

UNMAPPED is an AI-powered skills infrastructure that builds professional resilience. It allows workers to input their work history in plain language, which a Gemini-powered engine then deconstructs into verified skills and occupation matches anchored in the ESCO (European Skills, Competences, Qualifications, and Occupations) framework.

Beyond just a resume builder, the platform provides deep-layer labor market analysis. It assesses the user's specific automation risk—calibrated for their local economic context—and identifies high-growth "opportunity pathways." Users can download a portable PDF profile, browse global job postings, and apply for roles that match their verified AI-mapped skills.

How we built it

The platform is built on a modern, high-performance stack designed for scale. The frontend is a React application powered by Vite and Framer Motion, featuring a premium glassmorphic UI. For the backend, we utilized FastAPI to handle the complex AI processing and labor market calculations. Data persistence and user authentication are managed through a robust Supabase integration, replacing initial local storage with a centralized PostgreSQL database.

The core intelligence of UNMAPPED is driven by the Gemini 1.5 Flash model. We engineered a "Lean AI" architecture that offloads heavy natural language processing tasks—such as skill extraction and ISCO-08 occupation matching—to the cloud. This allowed us to keep the application responsive and lightweight without sacrificing the depth of the ESCO taxonomy.

Challenges we ran into

One of our biggest hurdles was the "5GB Problem." Initially, we attempted to use local machine learning libraries like Sentence-Transformers and PyTorch for skill matching. However, these libraries far exceeded Vercel’s 500MB serverless bundle limit. We had to pivot our entire architecture to a "Lean AI" model, refactoring the backend to offload these heavy tasks to the Gemini API while maintaining high accuracy.

Another significant challenge was the migration from client-side localStorage to a persistent Supabase database. We had to design a relational SQL schema that could handle the deeply nested JSON results produced by the AI—normalizing complex data across eight different tables while ensuring strict Row Level Security (RLS) for user privacy.

Accomplishments that we're proud of

We are incredibly proud of achieving a "Lean" deployment that packs 5GB of intelligence into a 100MB serverless footprint. Successfully navigating the constraints of cloud hosting forced us to be more creative with our architecture. We also take great pride in our PDF generation engine, which produces beautiful, standardized professional profiles that can be carried by workers anywhere in the world.

From a design perspective, we’re proud of creating an interface that feels premium and "state-of-the-art." We wanted workers in informal sectors to feel they are using a tool that is just as sophisticated and high-end as any Silicon Valley enterprise software, because their skills deserve that level of respect.

What we learned

Building UNMAPPED taught us the immense value of standardized labor taxonomies like ESCO and ISCO-08. We learned that the "secret sauce" of AI is not just in the models themselves, but in how you anchor their outputs to global, interoperable data standards. Without these anchors, AI results are just text; with them, they become a valid professional currency.

We also learned a valuable lesson in "serverless-first" development. Working within the hard limits of Vercel taught us to optimize our dependencies and rethink how we handle large datasets (like the 40MB ESCO CSV files), leading to a much faster and more cost-efficient application.

What's next for UNMAPPED

The next step for UNMAPPED is to build out a direct-hiring marketplace where employers can search specifically for "Verified AI-Mapped Skills." We want to move beyond just helping workers describe themselves and start directly facilitating the match between talent and opportunity.

We are also looking into mobile-first voice intake, allowing workers who may have lower literacy levels to simply "speak" their work history to the AI. Finally, we plan to expand our labor market signals to include more granular, rural-specific data, ensuring that UNMAPPED remains relevant for workers in even the most remote parts of the global economy.

Built With

• fastapi
• react