ApplyPilot

AI agents that research jobs, evaluate fit, prepare tailored applications, and keep humans in control.

Comment

Inspiration

Job searching should not feel like a second full-time job.

Today, candidates still have to move between multiple job boards, search for relevant positions, read and compare job descriptions, evaluate whether they are a good fit, and then prepare a resume, cover letter, and application answers for each opportunity.

The problem is not only the amount of information. Much of the work is repetitive.

We saw an opportunity to use AI not simply as another chatbot, but as an agent that can execute a job-search workflow step by step based on a goal defined by the user.

That is how ApplyPilot was born.

We imagined a simpler experience: a candidate should be able to say,

"This is the kind of job I want."

and let ApplyPilot execute the research and application-preparation work as a Mission, while the candidate remains in control of the final decision.

What it does

ApplyPilot is an AI-assisted job application operator that helps candidates run their job-search and application-preparation process in a more structured way.

The user first creates a Career Profile, which becomes the primary source of candidate information, including experience, target role, work preferences, expected salary, and skills.

The user then creates a Mission. A Mission is more than a simple job search. It gives the AI a clear objective describing the type of opportunities the candidate wants to find.

Once the Mission is launched, ApplyPilot runs several AI stages:

  1. Research Agent searches for relevant job opportunities using web search and Google Search Grounding.
  2. Matching Agent compares each opportunity against the candidate's Career Profile and produces a fit score, along with strengths and considerations.
  3. ApplyPilot presents the research and matching results in Mission Detail, allowing the candidate to see the discovered opportunities, their sources, job requirements, and why each position may or may not be a good fit.
  4. Once the workflow is complete, the Application Preparation Agent prepares a tailored application package, including a resume summary, cover letter, key talking points, and application Q&A.
  5. The candidate then performs a human review and decides whether to approve the prepared application.

The main output of ApplyPilot is therefore not simply an AI-generated response.

ApplyPilot produces two practical outcomes for the candidate:

  • Researched job opportunities with AI-evaluated candidate-to-job fit.
  • A tailored application package prepared for the selected opportunity.

ApplyPilot does not silently submit applications on behalf of the candidate. The candidate remains the final decision-maker.

How we built it

We built ApplyPilot as an agentic application workflow, rather than as one large AI prompt.

The core workflow is:

Career Profile → Mission → Research Agent → Matching Agent → Application Preparation → Human Review → Approval

Each stage has a distinct responsibility.

Career Profile represents the candidate's source of truth.

Mission defines the search objective.

Research Agent discovers job opportunities.

Matching Agent evaluates candidate-to-job fit.

Application Preparation Agent uses the candidate's verified information together with the target job context to prepare application materials.

The candidate then reviews the result before approving it.

For the AI workflow, ApplyPilot uses Gemini 3.5 Flash through the Gemini Developer API, integrated using the Google GenAI SDK.

The backend runs on Google Cloud Run. Application data such as Career Profiles, Missions, job Opportunities, Matches, and Applications are persisted in Google Cloud Firestore. The frontend is built with React and Vite and deployed through Firebase Hosting.

These Google technologies are not simply listed as project dependencies. Gemini and Google Cloud are actively used as part of the production workflow that runs ApplyPilot.

Technical requirements compliance

ApplyPilot directly satisfies the hackathon's required Google technology stack.

Gemini 3.5 or newer

ApplyPilot uses Gemini 3.5 Flash through the Gemini API as the model powering its AI workflow.

Gemini is used across the research, matching, and application-preparation stages.

Google Agent Framework

ApplyPilot uses the Google GenAI SDK to integrate Gemini-powered agent capabilities into the application workflow.

Gemini is not used merely as a standalone chatbot. Its capabilities are organized into multiple specialized stages, each with a specific responsibility within the ApplyPilot workflow.

Google Cloud infrastructure

ApplyPilot uses multiple Google Cloud services in its production application:

  • Google Cloud Run — runs the ApplyPilot backend.
  • Google Cloud Firestore — persists Career Profiles, Missions, Opportunities, Matches, and Applications.
  • Firebase Hosting — serves the production frontend.

These technologies are therefore not only present to satisfy a checklist. They are part of the actual production architecture and agent workflow of ApplyPilot.

Challenges we ran into

The hardest challenge was not simply connecting an LLM to job search.

The real challenge was helping candidates without allowing the AI to invent candidate experience.

Job descriptions often contain technologies and experience requirements that a candidate may not actually have. If those requirements are passed to an AI without clear boundaries, the model can easily produce a resume or cover letter that sounds impressive but incorrectly claims experience the candidate never had.

We did not want ApplyPilot to make candidates look more qualified by making things up.

So we designed an explicit separation:

Career Profile = candidate truth

Job Listing / Mission = target context

Job requirements must never automatically become candidate qualifications.

Another challenge was coordinating multiple AI stages while keeping the workflow understandable to the user. Research, matching, and application preparation each serve a different purpose, but they still need to feel like one continuous experience.

Accomplishments we're proud of

We are proud that ApplyPilot is not simply a demo that generates text from a prompt.

We built a working end-to-end agentic workflow that can be used from beginning to end.

A candidate can:

complete a Career Profile → create a Mission → let the agents research opportunities → review discovered jobs → see fit scores and considerations → receive a tailored application package → review the result → approve the application.

We are especially proud of the human-in-the-loop design.

AI performs the time-consuming work of researching, reading, comparing, and preparing application materials.

But AI does not make the final decision.

The candidate can review the result and decide whether the prepared application is appropriate to use.

Another accomplishment we are particularly proud of is factual grounding.

We implemented explicit rules to ensure that skills or requirements appearing only in a job listing are not automatically treated as the candidate's experience.

Most importantly, ApplyPilot is running as a real deployed production application, using Gemini and Google Cloud infrastructure.

What we learned

We learned that building useful AI agents is not simply about making a model generate better or longer text.

What matters more is giving AI a clear objective, dividing the work into appropriate stages, and establishing reliable boundaries between facts and context.

With ApplyPilot, we found that a simple separation between:

"What is true about the candidate?"

and

"What does the target job require?"

is extremely important.

The Career Profile provides candidate facts.

The job listing provides target context.

The Matching Agent can then use both to answer a much more useful question:

"How well does this opportunity fit this candidate, and why?"

We also learned that for tasks with real consequences, such as applying for a job, full automation is not always the best experience.

AI should reduce repetitive work without removing human control.

What's next for ApplyPilot

The current version of ApplyPilot focuses on the most repetitive parts of the job-search process: finding relevant opportunities, understanding candidate-to-job fit, and preparing applications.

The next step is to extend the agent workflow beyond the point where an application is approved.

We want ApplyPilot to eventually help candidates:

  • track application outcomes,
  • understand feedback from employers,
  • prepare for interviews,
  • improve their Career Profile based on real-world application results,
  • and use insights from previous applications to make future job searches more effective.

Our long-term vision is an AI career operator.

Not an AI that takes over someone's career, but an AI that handles the repetitive work behind the scenes so candidates can focus on the decisions that actually matter.

Find the opportunity. Understand the fit. Prepare the application. Keep the human in control.

Built With

  • firebase-hosting
  • gemini
  • gemini-api
  • google-cloud-firestore
  • google-cloud-run
  • google-genai-sdk
  • google-search-grounding
  • react
  • typescript
  • vite
Share this project:

Updates

Submission history