fix: progress logging, Ollama timeout for LLM extraction. Closes #132…

[utkarshqz]

fix: real-time progress logging and Ollama timeout for LLM extraction

Summary

Adds real-time console progress logging during LLM extraction and a request timeout to prevent indefinite hangs on slow or lower-end hardware.

The console was previously silent between [3] Starting extraction... and the final result — leaving developers unable to tell if the process was working or frozen. This PR fixes that with clear per-field [LOG] output and elapsed time reporting.

---

Closes / Fixes

Closes #132
Addresses #152

---

Type of change

• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)

---

What changed and why

1. 🪵 Real-time progress logging (src/llm.py)

Added [LOG] output immediately before the batch Ollama request so developers can see exactly which fields are being processed and confirm the backend is actively working.

Console output after this PR:

[LOG] Starting batch extraction for 7 field(s)...
[LOG] Queuing field 1/7 -> 'Name/SID'
[LOG] Queuing field 2/7 -> 'Job Title'
[LOG] Queuing field 3/7 -> 'Department'
[LOG] Queuing field 4/7 -> 'Phone Number'
[LOG] Queuing field 5/7 -> 'Email'
[LOG] Queuing field 6/7 -> 'Date'
[LOG] Queuing field 7/7 -> 'Signature'
[LOG] Sending single batch request to Ollama (model: mistral)...
[LOG] Ollama responded in 4.32s
[LOG] Batch extraction successful.

Fallback per-field extraction path also logs progress:

[LOG] Extracting field 1/7 -> 'Name/SID'
[LOG] Extracting field 2/7 -> 'Job Title'
...

 

2. ⏱️ Ollama request timeout (src/llm.py)

Added timeout=30 to requests.post().

Previously, if Ollama was slow or unresponsive, the process would hang indefinitely with no feedback. The 30-second timeout raises a clear requests.exceptions.Timeout which is caught and surfaced as a 503 error to the client — consistent with existing error handling.

This is especially important on lower-end hardware where CPU inference can be slow, as noted in #132.

---

How to test

# 1. Start Ollama
ollama serve

# 2. Start the API
uvicorn api.main:app --reload

# 3. Open frontend and fill a form
# OR hit the API directly:
curl -X POST http://localhost:8000/forms/fill \
  -H "Content-Type: application/json" \
  -d '{"template_id": 1, "input_text": "John Smith, firefighter..."}'

# 4. Watch the uvicorn console — you should see [LOG] lines appear
#    in real time during extraction

 

Run the test suite:

python -m pytest tests/ -v

Expected: 52 passed ✅

---

Screenshots / Evidence

Console output during extraction (verified locally):

[LOG] Starting batch extraction for 7 field(s)...
[LOG] Queuing field 1/7 -> 'Name/SID'
[LOG] Queuing field 2/7 -> 'Job Title'
[LOG] Queuing field 3/7 -> 'Department'
[LOG] Queuing field 4/7 -> 'Phone Number'
[LOG] Queuing field 5/7 -> 'Email'
[LOG] Queuing field 6/7 -> 'Date'
[LOG] Queuing field 7/7 -> 'Signature'
[LOG] Sending single batch request to Ollama (model: mistral)...
[LOG] Ollama responded in 4.32s
[LOG] Batch extraction successful.