ClearList

Inspiration

Job alerts are supposed to save time, but they often create a different problem: information overload. Sites like Indeed and LinkedIn can send dozens of listings every day, and many of them are jobs that the applicant is not actually qualified for.

I started wondering whether AI could act as a personal screening assistant. Instead of forcing users to read every listing themselves, could an AI review incoming job alerts and surface only the opportunities that are worth their time?

One challenge that particularly interested me was privacy. Since job alerts arrive through email, I needed a way for users to share listings with the application without forcing them to give up more access than they were comfortable with. To solve this, I designed two different ways for users to provide listings to ClearList.

What it does

ClearList automatically reviews job listings received from job alert emails and filters them based on a user's qualifications.

Users can choose between two privacy options:

Connect their inbox through a quick OAuth process.
Forward only job-related emails to a dedicated ClearList address so the application never sees unrelated messages.

Once listings are received, ClearList runs a two-tier AI review process. The first model performs fast filtering while the second model performs deeper qualification analysis. Jobs that pass both stages appear on a simple dashboard of opportunities worth the user's attention.

The goal is simple: spend less time reading irrelevant listings and more time applying to jobs that are actually a good fit.

How I built it

I initially used Claude to generate the foundation of both the frontend and backend. From there, I customized the application, configured the infrastructure, and deployed it.

The project uses:

Supabase for authentication, account management, email verification, and database storage
GitHub for version control
Vercel for hosting and deployment
OAuth integrations for inbox access
Encrypted token storage for security

For the AI pipeline, I used NVIDIA NIM's free API offerings.

Tier 1

Llama 3.1 8B Instruct
Fast preliminary filtering

Tier 2

Llama 3.3 70B Instruct
Detailed qualification analysis

The two-stage architecture reduces costs because only promising listings are sent to the larger model.

Challenges I ran into

The most difficult part of this project was infrastructure.

Configuring OAuth providers, authentication systems, environment variables, deployment pipelines, encryption keys, and API integrations took significantly longer than building the interface itself.

Another challenge was finding AI models that were powerful enough while still being affordable. Since I wanted to avoid paid APIs during development, I spent considerable time researching providers before eventually using NVIDIA NIM's free offerings.

Security was also a major focus. After generating the initial application, I opened a separate Claude conversation and asked it to analyze the website from an attacker's perspective. It identified several potential vulnerabilities, which I then addressed by improving authentication flows, token handling, and security practices.

Accomplishments that I'm proud of

Building a complete full-stack SaaS application
Designing a privacy-focused system with multiple data-sharing options
Creating a two-stage AI qualification pipeline
Learning OAuth authentication and secure credential management
Deploying a production web application using Supabase and Vercel
Conducting AI-assisted security reviews to strengthen the platform

What I learned

This project taught me that building software involves much more than writing code.

I learned how modern web applications manage authentication, securely store credentials, integrate third-party services, and deploy to cloud infrastructure. I also learned how important privacy and security considerations become when handling user data.

Most importantly, I learned that building a working prototype is only the first step. Making it secure, reliable, and scalable presents an entirely different set of challenges.

What's next for ClearList

Future improvements include:

Resume-aware qualification analysis
Support for additional job platforms
Better explanations for why jobs are accepted or rejected
Improved qualification scoring
Personalized recommendations
Scalability improvements for larger user bases

The long-term goal is to help job seekers focus on opportunities that genuinely match their experience instead of spending hours sorting through irrelevant listings.