500 error when workplace type is missing in job description

[LuisaG]

Describe the bug
A clear and concise description of what the bug is.

When submitting a job description that has an invalid or missing workplace type (e.g. 'N/A' instead of 'Remote', 'On-site', etc.), clicking the Improve button results in a 500 error. This shouldn't crash the app — the field should either be optional or safely default to 'Not Specified'.

If I manually add the workplace type in the job description example from above it works.

To Reproduce
Steps to reproduce the behavior:

1. Add a job description with 'N/A' or a missing work arrangement.

2. Click the Improve button.

3. Observe the 500 error.

Expected behavior

The app should validate or gracefully handle unexpected workplace type values. A missing or invalid value shouldn’t break functionality.

Screenshots
If applicable, add screenshots to help explain your problem.

Desktop (please complete the following information):

• OS: [e.g. iOS]
• Browser [e.g. chrome, safari]
• Version [e.g. 22]

Smartphone (please complete the following information):

• Device: [e.g. iPhone6]
• OS: [e.g. iOS8.1]
• Browser [e.g. stock browser, safari]
• Version [e.g. 22]

Additional context
The input is expected to be one of the following: 'Fully Remote', 'Hybrid', 'On-site', 'Remote', 'Not Specified', or 'Multiple Locations', but 'N/A' is being passed and isn’t supported. The system should sanitize or default this input instead of throwing a 500.

Edit: if I add the full link it works, as the location is found up top of the page.

[elliotclee]

Just ran into this too. I even tried putting "This job opening is in multiple locations." into the text of the job description and it still happened. I think the instructions to the LLM on how to convert a posting to JSON are not complete enough for it to know how to fill out that field.

[LuisaG]

Just ran into this too. I even tried putting "This job opening is in multiple locations." into the text of the job description and it still happened. I think the instructions to the LLM on how to convert a posting to JSON are not complete enough for it to know how to fill out that field.

Yes, but it shouldn't fail because of the location.

It's weird that it failed even with "multiple locations", maybe cause it's case sensitive? In the logs it says:

"Input should be 'Fully Remote', 'Hybrid', 'On-site', 'Remote', 'Not Specified' or 'Multiple Locations' [type=enum, input_value='N/A', input_type=str]"

Edit: it's case insensitive, found the code.

[Blindzord]

As a workaround for the issue, I used the following command from PowerShell:
ollama pull nomic-embed-text:137m-v1.5-fp16

[LuisaG]

As a workaround for the issue, I used the following command from PowerShell: ollama pull nomic-embed-text:137m-v1.5-fp16

Good to know! Wondering if the quality of the resume outputted is still the same with a different model.

[elliotclee]

As a workaround for the issue, I used the following command from PowerShell: ollama pull nomic-embed-text:137m-v1.5-fp16

That's a different bug with the same symptoms, I believe. The setup script doesn't 'ollama pull' the embeddings model like it should, but I already have the necessary embeddings model downloaded in ollama. The issue that LuisaG and I ran into had to do with it not parsing out the job location into JSON.

[Moosdijk]

As a workaround for the issue, I used the following command from PowerShell: ollama pull nomic-embed-text:137m-v1.5-fp16

The issue that LuisaG and I ran into had to do with it not parsing out the job location into JSON.

Improving the prompt helps somewhat as can be seen from #431 , but we need to figure out how to prompt the LLM to parse a wide variety of resume/information styles.

[LuisaG]

Improving the prompt helps somewhat as can be seen from #431 , but we need to figure out how to prompt the LLM to parse a wide variety of resume/information styles.

I addressed part of this in this PR, by expanding the enums for employment_type and remote_status to accept 'N/A', which was causing 500 errors when fields were missing or not aligned with expected values. This prevents hard crashes on imperfect job inputs and lets us prompt the LLM more safely.

You're right though — improving the prompt helps (as seen in #431), but we also need to make it more robust to different resume structures and edge cases. Happy to collaborate on refining that next!

[bpleonardo]

I think a better fix would be to specify the valid values in the schema presented to the AI. Just specifying "remote_status": "one of [...]" fixed the issue for me.
I explained it more in #447 (duplicate of this one)

[bpleonardo]

pydantic can generate json schemas, they would be a good approach if the AI understands them. The only thing I think AI could struggle with would be references, but we could run a simple script to inline them in the objects.

[LuisaG]

I think a better fix would be to specify the valid values in the schema presented to the AI. Just specifying "remote_status": "one of [...]" fixed the issue for me. I explained it more in #447 (duplicate of this one)

yes, good point! Not Specified and N/A are the same!