Analyze The Sentiment Of A Customer Call using LLM Gateway

1
import requests
2
import time
3
4
API_KEY = "YOUR_API_KEY"
5
audio_file_path = "./meeting.mp3"
6
7
# ------------------------------------------
8
# Step 1: Upload the audio file
9
# ------------------------------------------
10
def upload_file(filename):
11
    with open(filename, "rb") as f:
12
        upload_url = "https://api.assemblyai.com/v2/upload"
13
        headers = {"authorization": API_KEY}
14
        response = requests.post(upload_url, headers=headers, data=f)
15
        response.raise_for_status()
16
        return response.json()["upload_url"]
17
18
audio_url = upload_file(audio_file_path)
19
print(f"Uploaded audio file. URL: {audio_url}")
20
21
# ------------------------------------------
22
# Step 2: Request transcription
23
# ------------------------------------------
24
transcript_request = requests.post(
25
    "https://api.assemblyai.com/v2/transcript",
26
    headers={"authorization": API_KEY, "content-type": "application/json"},
27
    json={"audio_url": audio_url, "speech_models": ["universal-3-pro"]},
28
)
29
30
transcript_id = transcript_request.json()["id"]
31
32
# Poll until completed
33
while True:
34
    polling_response = requests.get(
35
        f"https://api.assemblyai.com/v2/transcript/{transcript_id}",
36
        headers={"authorization": API_KEY},
37
    )
38
    status = polling_response.json()["status"]
39
40
    if status == "completed":
41
        break
42
    elif status == "error":
43
        raise RuntimeError(f"Transcription failed: {polling_response.json()['error']}")
44
    else:
45
        print(f"Transcription status: {status}")
46
        time.sleep(3)
47
48
print("\nTranscription complete.\n")
49
50
# ------------------------------------------
51
# Step 3: Define questions
52
# ------------------------------------------
53
agent_context = "The agent is trying to get the customer to go through with the update to their car."
54
customer_context = "The customer is calling to check how much it would cost to update the map in his car."
55
56
answer_format = "<answer in one sentence> <reason in one sentence>"
57
58
questions = [
59
    {
60
        "question": "What was the overall sentiment of the call?",
61
        "context": customer_context,
62
        "answer_format": answer_format,
63
    },
64
    {
65
        "question": "What was the sentiment of the agent in this call?",
66
        "context": agent_context,
67
        "answer_format": answer_format,
68
    },
69
    {
70
        "question": "What was the sentiment of the customer in this call?",
71
        "context": customer_context,
72
        "answer_format": answer_format,
73
    },
74
    {
75
        "question": "What quote best demonstrates the customer's level of interest?",
76
        "context": customer_context,
77
        "answer_format": answer_format,
78
    },
79
    {
80
        "question": "Provide a quote from the agent that demonstrates their level of enthusiasm.",
81
        "context": agent_context,
82
        "answer_format": answer_format,
83
    },
84
]
85
86
# ------------------------------------------
87
# Step 4: Build prompt for the LLM
88
# ------------------------------------------
89
question_strs = []
90
for q in questions:
91
    q_str = f"Question: {q['question']}"
92
    if q.get("context"):
93
        q_str += f"\nContext: {q['context']}"
94
    if q.get("answer_format"):
95
        q_str += f"\nAnswer Format: {q['answer_format']}"
96
    question_strs.append(q_str)
97
98
questions_prompt = "\n\n".join(question_strs)
99
100
prompt = f"""
101
You are an expert at analyzing call transcripts.
102
Given the series of questions below, answer them accurately and concisely.
103
When context or answer format is provided, use it to guide your answers.
104
105
Transcript:
106
{{{{ transcript }}}}
107
108
Questions:
109
{questions_prompt}
110
"""
111
112
# ------------------------------------------
113
# Step 5: Query the LLM Gateway
114
# ------------------------------------------
115
headers = {"authorization": API_KEY}
116
117
response = requests.post(
118
    "https://llm-gateway.assemblyai.com/v1/chat/completions",
119
    headers=headers,
120
    json={
121
        "model": "claude-sonnet-4-5-20250929",
122
        "messages": [{"role": "user", "content": prompt}],
123
        "transcript_id": transcript_id,
124
        "max_tokens": 2000,
125
    },
126
)
127
128
response_json = response.json()
129
llm_output = response_json["choices"][0]["message"]["content"]
130
131
# ------------------------------------------
132
# Step 6: Parse and display the results
133
# ------------------------------------------
134
print("\n--- LLM Responses ---\n")
135
print(llm_output)

1
import requests
2
import time
3
4
API_KEY = "YOUR_API_KEY"

1
audio_file_path = "./meeting.mp3"
2
# ------------------------------------------
3
# Step 1: Upload the audio file
4
# ------------------------------------------
5
def upload_file(filename):
6
    with open(filename, "rb") as f:
7
        upload_url = "https://api.assemblyai.com/v2/upload"
8
        headers = {"authorization": API_KEY}
9
        response = requests.post(upload_url, headers=headers, data=f)
10
        response.raise_for_status()
11
        return response.json()["upload_url"]
12
audio_url = upload_file(audio_file_path)
13
print(f"Uploaded audio file. URL: {audio_url}")
14
# ------------------------------------------
15
# Step 2: Request transcription
16
# ------------------------------------------
17
transcript_request = requests.post(
18
    "https://api.assemblyai.com/v2/transcript",
19
    headers={"authorization": API_KEY, "content-type": "application/json"},
20
    json={"audio_url": audio_url, "speech_models": ["universal-3-pro"]},
21
)
22
transcript_id = transcript_request.json()["id"]
23
# Poll until completed
24
while True:
25
    polling_response = requests.get(
26
        f"https://api.assemblyai.com/v2/transcript/{transcript_id}",
27
        headers={"authorization": API_KEY},
28
    )
29
    status = polling_response.json()["status"]
30
    if status == "completed":
31
        break
32
    elif status == "error":
33
        raise RuntimeError(f"Transcription failed: {polling_response.json()['error']}")
34
    else:
35
        print(f"Transcription status: {status}")
36
        time.sleep(3)
37
print("\nTranscription complete.\n")

1
agent_context = "The agent is trying to get the customer to go through with the update to their car."
2
customer_context = "The customer is calling to check how much it would cost to update the map in his car."

1
answer_format = "<answer in one sentence> <reason in one sentence>"
2
questions = [
3
    {
4
        "question": "What was the overall sentiment of the call?",
5
        "context": customer_context,
6
        "answer_format": answer_format,
7
    },
8
    {
9
        "question": "What was the sentiment of the agent in this call?",
10
        "context": agent_context,
11
        "answer_format": answer_format,
12
    },
13
    {
14
        "question": "What was the sentiment of the customer in this call?",
15
        "context": customer_context,
16
        "answer_format": answer_format,
17
    },
18
    {
19
        "question": "What quote best demonstrates the customer's level of interest?",
20
        "context": customer_context,
21
        "answer_format": answer_format,
22
    },
23
    {
24
        "question": "Provide a quote from the agent that demonstrates their level of enthusiasm.",
25
        "context": agent_context,
26
        "answer_format": answer_format,
27
    },
28
]

1
# ------------------------------------------
2
# Step 4: Build prompt for the LLM
3
# ------------------------------------------
4
question_strs = []
5
for q in questions:
6
    q_str = f"Question: {q['question']}"
7
    if q.get("context"):
8
        q_str += f"\nContext: {q['context']}"
9
    if q.get("answer_format"):
10
        q_str += f"\nAnswer Format: {q['answer_format']}"
11
    question_strs.append(q_str)
12
questions_prompt = "\n\n".join(question_strs)
13
prompt = f"""
14
You are an expert at analyzing call transcripts.
15
Given the series of questions below, answer them accurately and concisely.
16
When context or answer format is provided, use it to guide your answers.
17
Transcript:
18
{{{{ transcript }}}}
19
Questions:
20
{questions_prompt}
21
"""

1
# ------------------------------------------
2
# Step 5: Query the LLM Gateway
3
# ------------------------------------------
4
headers = {"authorization": API_KEY}
5
response = requests.post(
6
    "https://llm-gateway.assemblyai.com/v1/chat/completions",
7
    headers=headers,
8
    json={
9
        "model": "claude-sonnet-4-5-20250929",
10
        "messages": [{"role": "user", "content": prompt}],
11
        "transcript_id": transcript_id,
12
        "max_tokens": 2000,
13
    },
14
)
15
response_json = response.json()
16
llm_output = response_json["choices"][0]["message"]["content"]
17
# ------------------------------------------
18
# Step 6: Parse and display the results
19
# ------------------------------------------
20
print("\n--- LLM Responses ---\n")
21
print(llm_output)