Extract Dialogue Data with LLM Gateway and JSON

1
import requests
2
import json
3
import os
4
import csv
5
import time
6
import re
7
8
# Configuration
9
api_key = "<YOUR_API_KEY>"
10
base_url = "https://api.assemblyai.com"
11
headers = {"authorization": api_key}
12
output_filename = "profiles.csv"
13
14
def extract_json(text):
15
    """Extract JSON from text that might contain markdown or extra text"""
16
    # First, try to remove markdown code blocks
17
    text = text.strip()
18
19
    # Remove ```json and ``` markers
20
    if text.startswith("```"):
21
        text = re.sub(r'^```(?:json)?\s*', '', text)
22
        text = re.sub(r'\s*```$', '', text)
23
24
    # Find the first { and last } to extract just the JSON object
25
    first_brace = text.find('{')
26
    last_brace = text.rfind('}')
27
28
    if first_brace != -1 and last_brace != -1:
29
        json_str = text[first_brace:last_brace + 1]
30
        return json.loads(json_str)
31
32
    # If that didn't work, try parsing the whole thing
33
    return json.loads(text)
34
35
def upload_file(file_path):
36
    """Upload a local audio file to AssemblyAI"""
37
    with open(file_path, "rb") as f:
38
        response = requests.post(f"{base_url}/v2/upload", headers=headers, data=f)
39
        if response.status_code != 200:
40
            print(f"Error uploading {file_path}: {response.status_code}, {response.text}")
41
            response.raise_for_status()
42
        return response.json()["upload_url"]
43
44
def transcribe_audio(audio_url):
45
    """Submit audio for transcription and poll until complete"""
46
    # Submit transcription request
47
    data = {"audio_url": audio_url, "speech_models": ["universal-3-pro"]}
48
    response = requests.post(f"{base_url}/v2/transcript", headers=headers, json=data)
49
50
    if response.status_code != 200:
51
        print(f"Error submitting transcription: {response.status_code}, {response.text}")
52
        response.raise_for_status()
53
54
    transcript_id = response.json()["id"]
55
    polling_endpoint = f"{base_url}/v2/transcript/{transcript_id}"
56
57
    # Poll for completion
58
    while True:
59
        transcript = requests.get(polling_endpoint, headers=headers).json()
60
        if transcript["status"] == "completed":
61
            return transcript_id
62
        elif transcript["status"] == "error":
63
            raise RuntimeError(f"Transcription failed: {transcript['error']}")
64
        else:
65
            time.sleep(3)
66
67
def process_with_llm_gateway(transcript_id, prompt):
68
    """Send transcript to LLM Gateway for processing"""
69
    llm_gateway_data = {
70
        "model": "claude-sonnet-4-5-20250929",
71
        "messages": [
72
            {
73
                "role": "user",
74
                "content": f"{prompt}\n\n{{{{ transcript }}}}"
75
            }
76
        ],
77
        "transcript_id": transcript_id,
78
        "max_tokens": 1500
79
    }
80
81
    response = requests.post(
82
        "https://llm-gateway.assemblyai.com/v1/chat/completions",
83
        headers=headers,
84
        json=llm_gateway_data
85
    )
86
87
    result = response.json()
88
89
    if "error" in result:
90
        raise RuntimeError(f"LLM Gateway error: {result['error']}")
91
92
    return result['choices'][0]['message']['content']
93
94
# Main execution
95
prompt = """
96
        You are an HR executive scanning through an interview transcript to extract information about a candidate.
97
        You are required to create a JSON response with key information about the candidate.
98
        You will use this template for your answer:
99
        {
100
            "Name": "<candidate-name>",
101
            "Position": "<job position that candidate is applying for>",
102
            "Past experience": "<A short phrase describing the candidate's relevant past experience for the role>"
103
        }
104
        Do not include any other text in your response. Only respond in JSON format that is not surrounded by markdown code, as your response will be parsed programmatically as JSON.
105
        """
106
107
# Get all files from interviews directory
108
interview_files = [os.path.join("interviews", file) for file in os.listdir("interviews")]
109
110
with open(output_filename, "w", newline="") as file:
111
    writer = csv.writer(file)
112
    header = ["Name", "Position", "Past Experience"]
113
    writer.writerow(header)
114
115
    print(f"Processing {len(interview_files)} interview files...")
116
117
    for interview_file in interview_files:
118
        print(f"\nProcessing: {interview_file}")
119
120
        # Upload file and get URL
121
        print("  Uploading file...")
122
        audio_url = upload_file(interview_file)
123
124
        # Transcribe audio
125
        print("  Transcribing...")
126
        transcript_id = transcribe_audio(audio_url)
127
128
        # Process with LLM Gateway
129
        print("  Analyzing with LLM Gateway...")
130
        llm_response = process_with_llm_gateway(transcript_id, prompt)
131
132
        # Parse JSON response
133
        interviewee_data = extract_json(llm_response)
134
        writer.writerow(interviewee_data.values())
135
        print(f"  Completed: {interviewee_data['Name']}")
136
137
print(f"\nCreated .csv file {output_filename}")

1
import requests
2
import json
3
import os
4
import csv
5
import time
6
import re
7
8
# Configuration
9
api_key = "<YOUR_API_KEY>"
10
base_url = "https://api.assemblyai.com"
11
headers = {"authorization": api_key}
12
output_filename = "profiles.csv"

1
def extract_json(text):
2
    """Extract JSON from text that might contain markdown or extra text"""
3
    # First, try to remove markdown code blocks
4
    text = text.strip()
5
6
    # Remove ```json and ``` markers
7
    if text.startswith("```"):
8
        text = re.sub(r'^```(?:json)?\s*', '', text)
9
        text = re.sub(r'\s*```$', '', text)
10
11
    # Find the first { and last } to extract just the JSON object
12
    first_brace = text.find('{')
13
    last_brace = text.rfind('}')
14
15
    if first_brace != -1 and last_brace != -1:
16
        json_str = text[first_brace:last_brace + 1]
17
        return json.loads(json_str)
18
19
    # If that didn't work, try parsing the whole thing
20
    return json.loads(text)

1
def upload_file(file_path):
2
    """Upload a local audio file to AssemblyAI"""
3
    with open(file_path, "rb") as f:
4
        response = requests.post(f"{base_url}/v2/upload", headers=headers, data=f)
5
        if response.status_code != 200:
6
            print(f"Error uploading {file_path}: {response.status_code}, {response.text}")
7
            response.raise_for_status()
8
        return response.json()["upload_url"]
9
10
def transcribe_audio(audio_url):
11
    """Submit audio for transcription and poll until complete"""
12
    # Submit transcription request
13
    data = {"audio_url": audio_url, "speech_models": ["universal-3-pro"]}
14
    response = requests.post(f"{base_url}/v2/transcript", headers=headers, json=data)
15
16
    if response.status_code != 200:
17
        print(f"Error submitting transcription: {response.status_code}, {response.text}")
18
        response.raise_for_status()
19
20
    transcript_id = response.json()["id"]
21
    polling_endpoint = f"{base_url}/v2/transcript/{transcript_id}"
22
23
    # Poll for completion
24
    while True:
25
        transcript = requests.get(polling_endpoint, headers=headers).json()
26
        if transcript["status"] == "completed":
27
            return transcript_id
28
        elif transcript["status"] == "error":
29
            raise RuntimeError(f"Transcription failed: {transcript['error']}")
30
        else:
31
            time.sleep(3)

1
def process_with_llm_gateway(transcript_id, prompt):
2
    """Send transcript to LLM Gateway for processing"""
3
    llm_gateway_data = {
4
        "model": "claude-sonnet-4-5-20250929",
5
        "messages": [
6
            {
7
                "role": "user",
8
                "content": f"{prompt}\n\n{{{{ transcript }}}}"
9
            }
10
        ],
11
        "transcript_id": transcript_id,
12
        "max_tokens": 1500
13
    }
14
15
    response = requests.post(
16
        "https://llm-gateway.assemblyai.com/v1/chat/completions",
17
        headers=headers,
18
        json=llm_gateway_data
19
    )
20
21
    result = response.json()
22
23
    if "error" in result:
24
        raise RuntimeError(f"LLM Gateway error: {result['error']}")
25
26
    return result['choices'][0]['message']['content']

1
prompt = """
2
        You are an HR executive scanning through an interview transcript to extract information about a candidate.
3
        You are required to create a JSON response with key information about the candidate.
4
        You will use this template for your answer:
5
        {
6
            "Name": "<candidate-name>",
7
            "Position": "<job position that candidate is applying for>",
8
            "Past experience": "<A short phrase describing the candidate's relevant past experience for the role>"
9
        }
10
        Do not include any other text in your response. Only respond in JSON format that is not surrounded by markdown code, as your response will be parsed programmatically as JSON.
11
        """

1
interview_files = [os.path.join("interviews", file) for file in os.listdir("interviews")]
2
3
with open(output_filename, "w", newline="") as file:
4
    writer = csv.writer(file)
5
    header = ["Name", "Position", "Past Experience"]
6
    writer.writerow(header)
7
8
    print(f"Processing {len(interview_files)} interview files...")
9
10
    for interview_file in interview_files:
11
        print(f"\nProcessing: {interview_file}")
12
13
        # Upload file and get URL
14
        print("  Uploading file...")
15
        audio_url = upload_file(interview_file)
16
17
        # Transcribe audio
18
        print("  Transcribing...")
19
        transcript_id = transcribe_audio(audio_url)
20
21
        # Process with LLM Gateway
22
        print("  Analyzing with LLM Gateway...")
23
        llm_response = process_with_llm_gateway(transcript_id, prompt)
24
25
        # Parse JSON response
26
        interviewee_data = extract_json(llm_response)
27
        writer.writerow(interviewee_data.values())
28
        print(f"  Completed: {interviewee_data['Name']}")
29
30
print(f"\nCreated .csv file {output_filename}")