Extract Transcript Quotes with LLM Gateway

1
import requests
2
import json
3
import time
4
5
API_KEY = "YOUR_API_KEY"
6
headers = {"authorization": API_KEY}
7
8
# Transcribe the audio file
9
print("Submitting audio for transcription...")
10
transcript_response = requests.post(
11
    "https://api.assemblyai.com/v2/transcript",
12
    headers=headers,
13
    json={
14
        "audio_url": "https://assembly.ai/wildfires.mp3",
15
        "speaker_labels": True
16
    }
17
)
18
19
transcript_id = transcript_response.json()["id"]
20
21
# Poll for transcription completion
22
while True:
23
    transcript_result = requests.get(
24
        f"https://api.assemblyai.com/v2/transcript/{transcript_id}",
25
        headers=headers
26
    ).json()
27
    
28
    if transcript_result["status"] == "completed":
29
        break
30
    elif transcript_result["status"] == "error":
31
        raise Exception(f"Transcription failed: {transcript_result['error']}")
32
    
33
    time.sleep(3)
34
35
# Extract utterances with timestamps
36
utterances_data = [
37
    {"text": u["text"], "start": u["start"], "end": u["end"], "speaker": u["speaker"]} 
38
    for u in transcript_result["utterances"]
39
]
40
41
# Create prompt with timestamped utterances
42
prompt = f"""You are analyzing a transcript with timestamped utterances. Each utterance includes the text content, speaker label, and start/end timestamps in milliseconds.
43
44
Here is the transcript data:
45
{json.dumps(utterances_data, indent=2)}
46
47
Task: Identify the 3-5 most engaging, impactful, or quotable utterances from this transcript.
48
49
Return your response as a JSON array with the following structure:
50
{{
51
  "quotes": [
52
    {{
53
      "text": "exact quote text",
54
      "start": start_timestamp_in_milliseconds,
55
      "end": end_timestamp_in_milliseconds,
56
      "speaker": "speaker_label",
57
      "reason": "brief explanation of why this quote is engaging"
58
    }}
59
  ]
60
}}
61
62
Return ONLY valid JSON, no additional text."""
63
64
# Use LLM Gateway to extract quotes
65
print("Submitting transcript to LLM Gateway for quote extraction...")
66
gateway_response = requests.post(
67
    "https://llm-gateway.assemblyai.com/v1/chat/completions",
68
    headers=headers,
69
    json={
70
        "model": "gpt-5-nano",
71
        "messages": [
72
            {"role": "user", "content": prompt}
73
        ]
74
    }
75
)
76
77
result = gateway_response.json()
78
quotes_json = json.loads(result["choices"][0]["message"]["content"])
79
print(json.dumps(quotes_json, indent=2))

1
import requests
2
import json
3
import time
4
5
API_KEY = "YOUR_API_KEY"
6
headers = {"authorization": API_KEY}

1
# Transcribe the audio file
2
print("Submitting audio for transcription...")
3
transcript_response = requests.post(
4
    "https://api.assemblyai.com/v2/transcript",
5
    headers=headers,
6
    json={
7
        "audio_url": "https://assembly.ai/wildfires.mp3",
8
        "speaker_labels": True
9
    }
10
)
11
12
transcript_id = transcript_response.json()["id"]
13
14
# Poll for transcription completion
15
while True:
16
    transcript_result = requests.get(
17
        f"https://api.assemblyai.com/v2/transcript/{transcript_id}",
18
        headers=headers
19
    ).json()
20
    
21
    if transcript_result["status"] == "completed":
22
        break
23
    elif transcript_result["status"] == "error":
24
        raise Exception(f"Transcription failed: {transcript_result['error']}")
25
    
26
    time.sleep(3)

1
utterances_data = [
2
    {"text": u["text"], "start": u["start"], "end": u["end"], "speaker": u["speaker"]} 
3
    for u in transcript_result["utterances"]
4
]

1
# Create prompt with timestamped utterances
2
prompt = f"""You are analyzing a transcript with timestamped utterances. Each utterance includes the text content, speaker label, and start/end timestamps in milliseconds.
3
4
Here is the transcript data:
5
{json.dumps(utterances_data, indent=2)}
6
7
Task: Identify the 3-5 most engaging, impactful, or quotable utterances from this transcript.
8
9
Return your response as a JSON array with the following structure:
10
{{
11
  "quotes": [
12
    {{
13
      "text": "exact quote text",
14
      "start": start_timestamp_in_milliseconds,
15
      "end": end_timestamp_in_milliseconds,
16
      "speaker": "speaker_label",
17
      "reason": "brief explanation of why this quote is engaging"
18
    }}
19
  ]
20
}}
21
22
Return ONLY valid JSON, no additional text."""
23
24
# Use LLM Gateway to extract quotes
25
print("Submitting transcript to LLM Gateway for quote extraction...")
26
gateway_response = requests.post(
27
    "https://llm-gateway.assemblyai.com/v1/chat/completions",
28
    headers=headers,
29
    json={
30
        "model": "gpt-5-nano",
31
        "messages": [
32
            {"role": "user", "content": prompt}
33
        ]
34
    }
35
)
36
37
result = gateway_response.json()
38
quotes_json = json.loads(result["choices"][0]["message"]["content"])
39
print(json.dumps(quotes_json, indent=2))