JobJam

Real People. Real Practice. Real Growth.

Comment

About JobJam: Technical Overview

Motivation

Career prep tools often focus on AI-driven feedback but lack human connection. JobJam bridges this gap by connecting users with similar goals for real-time, anonymous video calls, allowing them to practice networking, interviews, and public speaking with actual humans rather than just AI simulations.

System Architecture

JobJam is built using:

  • Frontend: SwiftUI
  • Backend & Database: Firebase Authentication & Firestore
  • Video Calls: Agora with token server hosted on Render
  • AI & Matching: Hugging Face model intfloat/e5-large-v2 for embeddings, Gemini for JobBot
  • Matching Algorithm: Cosine similarity of embeddings

The system workflow:

  1. User Authentication: Users sign in anonymously via Firebase.
  2. Input Processing: Users enter a job description (d).
  3. Embedding Generation: The description is converted into a vector embedding (\mathbf{v}_d \in \mathbb{R}^n) using the Hugging Face model.
  4. Storage: The embedding is stored in Firestore and added to a matchmaking queue.
  5. Similarity Matching: Users are matched based on cosine similarity.

Cosine Similarity

To match users, we compute the cosine similarity between embeddings:

[ \text{similarity}(\mathbf{v}_i, \mathbf{v}_j) = \frac{\mathbf{v}_i \cdot \mathbf{v}_j}{|\mathbf{v}_i| |\mathbf{v}_j|} ]

Where:

  • (\mathbf{v}_i, \mathbf{v}_j \in \mathbb{R}^n) are the embeddings of two users’ job descriptions.
  • (\mathbf{v}_i \cdot \mathbf{v}_j) is the dot product of the two vectors.
  • (|\mathbf{v}_i|) is the Euclidean norm of (\mathbf{v}_i).

A pair is matched if:

[ \text{similarity}(\mathbf{v}_i, \mathbf{v}_j) \geq \theta ]

where (\theta) is a tunable threshold representing how closely aligned two users’ career interests are.

Real-Time Video Integration

Once a match is found:

  • Token Generation: The app requests a temporary Agora token from a server hosted on Render.
  • Video Call Setup: Users are connected in a peer-to-peer or group call depending on future features.
  • In-Call Features: Mute/unmute audio, disable video, leave call, access JobBot AI assistant.

AI Assistant: JobBot

JobBot is powered by Gemini and allows users to:

  • Ask career-related questions in real time.
  • Practice interview prompts during calls.
  • Receive instant feedback or suggestions.

Challenges & Learnings

  • Designing anonymous yet meaningful matchmaking.
  • Ensuring low-latency vector comparisons for real-time pairing.
  • Handling asynchronous API calls while maintaining a smooth user experience.
  • Integrating video streams and AI services simultaneously without blocking the UI.

Next Steps

  • Local Audio Recording: Capture user audio during calls.
  • Transcription & Evaluation: Use AssemblyAI for post-call analysis.
  • Personalization: Improve JobBot with tailored prompts and suggestions.
  • Scalability: Optimize matchmaking for large queues while maintaining speed.

Built With

Share this project:

Updates