Melody: AI Career Matchmaker

Melody is a Gemini Live Agent who "sees" your resume to get a sense of who you are, "speaks" to you about your goals and "listens" to your replies to dig deeper than a basic search and find a match.

Comment

THE PROBLEM

Searching for a job is an emotional process. The current tools pretend it isn't.

Google Jobs: sterile, linear, keyword-first. One search at a time. No sentiment. No memory. No sense of who you actually are.

LinkedIn: closer (profile matching, personalized recs) but still abstracts the person away into a social media profile and assumes that an account is an adequate substitute for a human being.

The result: Users apply into a void. No signal. No momentum. Creeping self-doubt. Declining self-esteem. The emotional cost of job searching is enormous and no product is designed around it.

The current paradigm also assumes that what matters most to users is job title, salary, and location. In reality, people leave jobs and seek new ones for deeply personal, emotional, and relational reasons, and yet no tool lets them say so.

THE INSIGHT

People are not meant to face a void.

Today's job seekers don't fit the typical cast of characters: the frantic new grad, the laid-off white collar worker shaking her fist at "AI." They are all kinds of people. What they share is this: they are human, and humans need the void to talk back.

Not in three weeks when a recruiter calls. Right away to give them momentum. To give them a feeling of collaboration.

People yearned for BuzzFeed quizzes and compatibility calculators where you'd enter your name and your crush's name. It never mattered how accurate it was. It mattered that something knew about you, about your possibilities, about what could be. It was fun. It was validating. It was human.

Melody is that but real, and maybe more useful.

THE PRODUCT

Melody is a live voice agent and AI job search matchmaker. She listens to how you talk about your work, not just what your resume says, and finds jobs that match who you actually are.

She uses emotional intelligence to get to the root of what matters to you in your next role:

  • Hears the hesitation when you describe why you left
  • Catches the energy when you describe what you love to build
  • Flags contradictions ("you said you hate micromanagement but also that you want more structure. Let's figure that out")
  • Translates resume language back into human language: "Led cross-functional initiatives" becomes "you want ownership without being a people manager"

Her goal is not to get you a job. Her goal is to help you understand yourself well enough that the right job becomes obvious and to save you the time and emotional cost of pursuing the wrong ones.

THE USERS

Primary Users

People for whom the textbox paradigm is a genuine barrier, not just an inconvenience:

  • People with disabilities for whom typing is inaccessible
  • Anxious users who spiral when left alone with a search bar
  • Oral/audio learners who are better at explaining themselves in conversation than in written keywords
  • Social/collaborative thinkers who soundboard rather than think alone
  • High-standards users who expect modern, customized experiences and find current tools too impersonal

When They Reach for Melody

Not when their job search is going well. When they're demoralized. Frustrated. Beginning to doubt themselves. Tired of the void.

Their Motivations (in order)

  1. Financial: they need a job and they'll try new tech to get it
  2. Emotional: once they use it, they come back for momentum and to think outside whatever box they have gotten stuck in

Edge Case / Guardrail

Melody must recognize over-reliance. If a user shares deeply personal details repeatedly or uses the app compulsively, Melody gently redirects them to human support like a therapist or a career coach, and surfaces a shortlist of actionable first steps so the conversation has somewhere real to go.

WHAT MELODY DOES (FEATURE SET)

Listen & Interpret

  • Live bidi voice conversation (Gemini Live API)
  • Extracts must-haves, nice-to-haves, and absolute-nots from natural speech instead of bullet points
  • Flags and surfaces contradictions for the user to resolve before they become search criteria
  • Translates resume lines into the sentiment behind them
  • Builds a dynamic priority model in real time

Search & Filter

  • Google Search → live job listings
  • Parses job descriptions: title/location first, then filters description text against the user's stated desires and absolute-nots, not just keyword matching
  • Callbacks enforce source quality and result relevance

Remember & Refine

  • Vertex AI MemoryBank (Public Preview) persists extracted facts across sessions, scoped per user
  • PreloadMemoryTool in Melody's tools list injects relevant memories at the start of each new session automatically
  • After session close, memories.generate() extracts and stores what's valuable. Melody configures custom topics: USER_PREFERENCES, CONVERSATION_EVENTS, job_search_state
  • "Last time you said you hated micromanagement and wanted a collaborative team" is a stored memory, not a session variable
  • Users don't have to repeat themselves across sessions; Melody picks up where she left off
  • Memory generation is not automatic and must be triggered explicitly at session close

Explain & Recommend

  • When Melody has gathered enough signal across a few turns, she triggers the search mid-conversation so the user doesn't have to ask
  • Silently calls google_search , builds JobMatchReport from tool results, saves it to file
  • Reads the matches back to the user in voice: role, company, why she picked it
  • Closes every session with a send-off: wishes the user good luck and reminds them what she learned about them
  • JobMatchReport schema:
    • role, company, culture_fit_score
    • why_melody_picked_this (in plain English, read aloud)
    • apply_url
Share this project:

Updates

Submission history