Looking for Your Valuable Feedback & Suggestions

#1
by mananshah296 - opened
Agents-MCP-Hackathon org

Hey everyone,

I’ve been working on Career Compass AI for the Agent Demo Track 3 — it’s a hybrid-agent (agent + native LLM) tool that helps you:

  1. Search for jobs smartly
  2. Match them with your resume
  3. And generate cover letters based on context

I’d really love your honest feedback — anything that can help me make it better!

Some things I’d love your thoughts on:

  1. Did the job/resume matching feel accurate or helpful?
  2. Was the cover letter generator actually useful?
  3. Any bugs, slowdowns, or UI quirks that stood out?
  4. Did the basic vs. advanced search make sense?

Feature-wise:

  1. Anything you wish it could do that it currently doesn’t?
  2. Ideas for extra tools, better analysis, or integrations?

Feel free to be as honest or detailed as you want!
Thanks a ton, and good luck on all your builds too...

— Manan

Agents-MCP-Hackathon org

Hi 👋,

Great idea and a clean implementation — I especially like the solid documentation and architecture! 👏

Just a heads-up: it looks like your secret key (eyJhbGciO...) might be exposed in the frontend. You’re using value=os.environ.get("NEBIUS_API_KEY", ""), but double-check that it isn’t being passed to the client unintentionally. Secrets should stay server-side to avoid potential security risks 🔐

Also, are you leveraging Google Jobs in the backend? That part caught my curiosity — would love to hear more about the stack you're using.

Keep it up — really promising work! 🚀
Cheers,
Chris

Agents-MCP-Hackathon org
edited 1 day ago

Hey Chris,

Thanks so much for the detailed feedback — I really appreciate it!

And good catch on the secret key. I did set it as a private secret on HF so users wouldn’t have to enter it (since we had free Nebius credits), but I didn’t realize it might still be exposed on the frontend. I’ll definitely dig into that. Do you have any tips on ensuring it's fully server-side?

Also, great point about the job search via SerpAPI. I actually tried a few approaches before that — using DuckDuckGo with a browsing tool worked for finding jobs, but the agent often hallucinated the job apply links. I also tried scraping (even with Crawl4AI), but many job sites blocked it. With limited time, SerpAPI turned out to be the most reliable option for accurate job links.

I have attached all the details along with demo links and short video here: https://www.linkedin.com/posts/manan-j-shah_ai-hackathon-jobsearch-activity-7339341523957821441-6N_B?utm_source=share&utm_medium=member_desktop&rcm=ACoAACxJaEoBCCmQ-l2XSrWEHKhy3p45-5-LpQg

Thanks again — this kind of feedback is super helpful!

Agents-MCP-Hackathon org

Hi

And good catch on the secret key. I did set it as a private secret on HF so users wouldn’t have to enter it (since we had free Nebius credits), but I didn’t realize it might still be exposed on the frontend. I’ll definitely dig into that. Do you have any tips on ensuring it's fully server-side?

Instead of putting os.environ.get("NEBIUS_API_KEY", "") in gr.Textbox
serp_api_key = gr.Textbox(
label="SerpAPI Key",
placeholder="Enter your SerpAPI key here...",
type="password",
value=os.environ.get("SERP_API_KEY", ""),
info="Required for all job searches",
elem_classes=["api-input"]
)
nebius_api_key = gr.Textbox(
label="Nebius API Key",
placeholder="Enter your Nebius API key here...",
type="password",
value=os.environ.get("NEBIUS_API_KEY", ""),
info="Required for AI-enhanced searches",
elem_classes=["api-input"]
)

rather put it in a function that is called later or in a config file.

Chris

Sign up or log in to comment