• 

  MimicIn Logo

• 

  ADK system

MimicIn

Inspiration

We wanted to create a way for students and job seekers to practice realistic conversations with recruiters, employees, and executives without the pressure of a live interaction. The inspiration came from coffee chats and career fairs, where preparation and confidence often determine success.

What it does

MimicIn is a multi-agent system that simulates professional coffee chats. A Persona agent generates context-aware prompts, while Converse agents act as recruiters, employees, or executives. Users can upload or scrape LinkedIn profiles, and the system builds tailored personas that respond dynamically in conversation. This helps users practice asking thoughtful questions, navigating dialogue, and building confidence.

How we built it

• Google ADK (Agent Development Kit): Used for multi-agent orchestration and agent-to-agent (A2A) communication.
  • Persona Agent: Takes LinkedIn profile links as input and generates a character persona based on the contents of the profile. Uses workflows and tools across multiple sub-agents to output a final natural language prompt for use.
    • Selenium Web Scraping w/ Gemini parsing
    • Google Search for additional research
  • Converse Agent: Performs real-time bi-directional (BiDi) conversations with the end user. Imitates the provided persona, and has access to Google Search grounding for access to additional relevant information.
    • Unity frontend with custom model and UI logic.
    • FastAPI backend server to handle agent<->client communication.
    • Real-time websockets with custom audio handling.
    • Google Search for grounding

Challenges we ran into

• LinkedIn has very aggressive authentication walls for most profiles, making it difficult to fetch the necessary data.
  • This ultimately caused our Persona agent to remain incomplete, and we currently use a static profile and prompt in place of it.
• Maintaining session state throughout the pipeline as it's passed through multiple agents.
• Use of unfamiliar tools, like Unity, the Google ADK, and websockets created confusion at multiple points of the development.

Accomplishments that we're proud of

• A custom agent that simulates a realistic conversation with a recruiter.
• A custom-made BiDi implementation in Unity, with audio buffers and PCM-16 conversions.
• A ambitious design of agents, with an impressive workflow being maintained across them.
• Building something that can genuinely help students and professionals prepare for networking.

What we learned

• Use of the Google ADK to orchestrate complex agent pipelines.
• Best practices for Unity version control.
• The challenges that web scraping faces, and the possible solutions to them.

What's next for MimicIn

• Enhancing the frontend for a better user experience.
• A finalized implementation of the Persona agent's LinkedIn scraping.
• Expanding persona generation to include company values, recent news, or technical interview prep.
• Use of virtual avatars, lip-syncing, and NLP technology to create a more reactive and engaging conversational experience.

Built With

• a2a
• adk
• c#
• python
• unity