Inspiration

Too many people are ghosted from job applications. Start-ups are unable to afford a huge HR team to filer through applicants quickly. And large company HR are too alienised from the work to get a clear understanding of the technical needs. So we created a service that matches a Company Job posting to an Technical Applicant resume and provide full disclosure with reasoning on the results. So applicant know what to do better, and companies have a more informed decision to make.

What it does

The basics is that we take in the companies job posting then elevate the technical side by using AI with the added information of the company's internal or public GitHub to ensure even terribly written postings can still be evaluated. Then we take in the applicants resume analyse it with AI and if a GitHub provided analyse the GitHub then judge how well the candidate fits the role. Providing a full transparent reasoning to both the applicant and the employer. This is done while ensuring the AI is being ethical and not adhering to bias as much as possible.

How we built it

By using two pipelines for our agents we separated the act of the AI elevating bad job descriptions and evaluating resumes and returns with suggestion if any. We also used Toon to reduce the token as well. By using ethical prompting techniques we ensure that all resumes are judged fairly. We also provide a clean front end to access the app which also tell the employers and the candidates why or why they were not suited to the job. Giving full transparency of how the AI deduced its answer.

Challenges:

Ensuring ethical, bias-aware reasoning while keeping the system fast and token-efficient was also a major technical challenge.

Accmplishments

We built a clean, intuitive front end that visualizes the entire multi-agent hiring pipeline, including branching logic, evaluation outputs, and match scores. Integrating real-time agent results into an interactive flow diagram was a major technical win that made the system transparent and easy to understand.

What we learned

We learnt about how to implement AI in a responsible way. Which pertains to the use of models and telemetry data to ensure AI always gives a similar response every time reducing the chance of hallucinations. Reducing the use of tokens and making AI prompts more efficient and viable to use was also important in the development of our solution. We also researched and learnt how to manipulate prompts to ensure the AI is more ethical, by favouring or un-favouring certain paths.

What's next for Synapse

Synapse has now implemented a basic AI checker for Resume compared to job description. Next we plan to implement a solution to create job descriptions from companies by checking with different people in the hierarchy and similar job roles across the industry to ensure that the company is actually hiring a person that it needs and afford. This makes the job market more unsaturated as requirement won't be as high.