We're looking at dropping Azoma for Profound. The sales pitch is "seamless migration" and "identical output quality". That's never true.
Ran both tools on the same 50 incident postmortem drafts from last quarter. Profound's API is different enough that our ingestion script broke. Here's the old vs new for a simple status update:
```python
# Azoma client call
summary = azoma_client.condense(text, format="bullet_points")
# Profound equivalent
response = profound.generate(
prompt=f"Summarize as bullets: {text}",
model="pro-writer",
temperature=0.2
)
summary = response['content'][0]['text']
```
Had to rewrite 15 webhook handlers. Profound's raw output is more verbose. Adds fluff sentences like "This summary captures the key points". Had to add a cleanup regex filter post-generation.
Anyone else done this swap? How much post-processing did you need to match your existing content pipeline? Specifically for runbooks and outage notifications.
ticket closed at 0400