Loan Advisory Agent using Pipecat

1
import os
2
from dotenv import load_dotenv
3
from loguru import logger
4
from pipecat.frames.frames import LLMRunFrame
5
from pipecat.pipeline.pipeline import Pipeline
6
from pipecat.pipeline.runner import PipelineRunner
7
from pipecat.pipeline.task import PipelineTask
8
from pipecat.processors.aggregators.llm_context import LLMContext
9
from pipecat.processors.aggregators.llm_response_universal import (
10
    LLMContextAggregatorPair,
11
)
12
from pipecat.runner.types import RunnerArguments
13
from pipecat.runner.utils import create_transport
14
from pipecat.services.sarvam.stt import SarvamSTTService
15
from pipecat.services.sarvam.tts import SarvamTTSService
16
from pipecat.services.openai.llm import OpenAILLMService
17
from pipecat.transports.base_transport import TransportParams
18
from pipecat.transports.daily.transport import DailyParams
19
20
load_dotenv(override=True)
21
22
async def bot(runner_args: RunnerArguments):
23
    """Main bot entry point."""
24
    
25
    # Create transport (supports both Daily and WebRTC)
26
    transport = await create_transport(
27
        runner_args,
28
        {
29
            "daily": lambda: DailyParams(audio_in_enabled=True, audio_out_enabled=True),
30
            "webrtc": lambda: TransportParams(
31
                audio_in_enabled=True, audio_out_enabled=True
32
            ),
33
        },
34
    )
35
36
    # Initialize AI services
37
    stt = SarvamSTTService(
38
        api_key=os.getenv("SARVAM_API_KEY"),
39
        language="unknown",  # Auto-detect for diverse customer base
40
        model="saarika:v2.5"
41
    )
42
    
43
    tts = SarvamTTSService(
44
        api_key=os.getenv("SARVAM_API_KEY"),
45
        target_language_code="en-IN",
46
        model="bulbul:v2",
47
        speaker="abhilash"  # Professional and trustworthy voice
48
    )
49
    
50
    llm = OpenAILLMService(api_key=os.getenv("OPENAI_API_KEY"), model="gpt-4o")
51
52
    # Set up conversation context with loan advisor personality
53
    messages = [
54
        {
55
            "role": "system",
56
            "content": """You are a professional and helpful loan advisory agent for a leading financial institution in India.
57
58
Your expertise covers:
59
60
**Loan Products:**
61
- Personal Loans: Unsecured loans for various personal needs
62
- Home Loans: For purchasing, constructing, or renovating homes
63
- Vehicle Loans: Car loans, two-wheeler loans
64
- Business Loans: For SMEs, startups, and working capital
65
- Education Loans: For domestic and international education
66
- Gold Loans: Secured loans against gold jewelry
67
- Loan Against Property (LAP): Secured loans using property as collateral
68
69
**Key Information to Provide:**
70
- Interest rates (mention that actual rates depend on profile and market conditions)
71
- Typical loan amounts and tenure options
72
- Eligibility criteria (age, income, credit score, employment type)
73
- Required documents (ID proof, address proof, income documents, etc.)
74
- Processing fees and other charges
75
- EMI calculation basics
76
- Prepayment and foreclosure options
77
78
**Communication Guidelines:**
79
- Be professional, trustworthy, and helpful
80
- Explain financial terms in simple language
81
- Never guarantee loan approval - mention that final approval depends on detailed assessment
82
- Always recommend customers to read terms and conditions carefully
83
- If asked about specific interest rates, provide indicative ranges and mention they vary
84
- Encourage customers to visit the branch or website for exact current rates
85
- Be transparent about fees and charges
86
- Help customers understand their EMI burden before recommending loan amounts
87
- Suggest customers maintain a good credit score
88
89
**Compliance Reminders:**
90
- Never make false promises about loan approval
91
- Always mention that loans are subject to eligibility and documentation
92
- Recommend customers to compare offers before deciding
93
- Remind about the importance of timely repayments
94
95
Start by greeting the customer professionally and asking how you can help them with their loan requirements.""",
96
        },
97
    ]
98
    context = LLMContext(messages)
99
    context_aggregator = LLMContextAggregatorPair(context)
100
101
    # Build pipeline
102
    pipeline = Pipeline(
103
        [
104
            transport.input(),
105
            stt,
106
            context_aggregator.user(),
107
            llm,
108
            tts,
109
            transport.output(),
110
            context_aggregator.assistant(),
111
        ]
112
    )
113
114
    task = PipelineTask(pipeline)
115
116
    @transport.event_handler("on_client_connected")
117
    async def on_client_connected(transport, client):
118
        logger.info("Customer connected")
119
        messages.append(
120
            {"role": "system", "content": "Greet the customer professionally and ask how you can help them with their loan requirements."}
121
        )
122
        await task.queue_frames([LLMRunFrame()])
123
124
    @transport.event_handler("on_client_disconnected")
125
    async def on_client_disconnected(transport, client):
126
        logger.info("Customer disconnected")
127
        await task.cancel()
128
129
    runner = PipelineRunner(handle_sigint=runner_args.handle_sigint)
130
    await runner.run(task)
131
132
if __name__ == "__main__":
133
    from pipecat.runner.run import main
134
    main()

1
stt = SarvamSTTService(
2
    api_key=os.getenv("SARVAM_API_KEY"),
3
    language="hi-IN",  # Hindi
4
    model="saarika:v2.5"
5
)
6
7
tts = SarvamTTSService(
8
    api_key=os.getenv("SARVAM_API_KEY"),
9
    target_language_code="hi-IN",
10
    model="bulbul:v2",
11
    speaker="karun"  # Professional male voice
12
)
13
14
llm = OpenAILLMService(api_key=os.getenv("OPENAI_API_KEY"), model="gpt-4o")

1
stt = SarvamSTTService(
2
    api_key=os.getenv("SARVAM_API_KEY"),
3
    language="ta-IN",
4
    model="saarika:v2.5"
5
)
6
7
tts = SarvamTTSService(
8
    api_key=os.getenv("SARVAM_API_KEY"),
9
    target_language_code="ta-IN",
10
    model="bulbul:v2",
11
    speaker="anushka"
12
)
13
14
llm = OpenAILLMService(api_key=os.getenv("OPENAI_API_KEY"), model="gpt-4o")

1
stt = SarvamSTTService(
2
    api_key=os.getenv("SARVAM_API_KEY"),
3
    language="unknown",  # Auto-detects language
4
    model="saarika:v2.5"
5
)
6
7
tts = SarvamTTSService(
8
    api_key=os.getenv("SARVAM_API_KEY"),
9
    target_language_code="en-IN",
10
    model="bulbul:v2",
11
    speaker="abhilash"
12
)
13
14
llm = OpenAILLMService(api_key=os.getenv("OPENAI_API_KEY"), model="gpt-4o")

1
# Customer speaks Hindi/Tamil/etc. → Saaras converts to English → LLM processes
2
3
stt = SarvamSTTService(
4
    api_key=os.getenv("SARVAM_API_KEY"),
5
    model="saaras:v2.5"  # Speech-to-English translation
6
)
7
8
tts = SarvamTTSService(
9
    api_key=os.getenv("SARVAM_API_KEY"),
10
    target_language_code="en-IN",
11
    model="bulbul:v2",
12
    speaker="abhilash"
13
)
14
15
llm = OpenAILLMService(api_key=os.getenv("OPENAI_API_KEY"), model="gpt-4o")

1
tts = SarvamTTSService(
2
    api_key=os.getenv("SARVAM_API_KEY"),
3
    target_language_code="en-IN",
4
    model="bulbul:v2",
5
    speaker="abhilash",
6
    pitch=0.0,           # Range: -1.0 to 1.0
7
    pace=1.0,            # Normal pace for clear communication
8
    loudness=1.5,        # Range: 0.5 to 2.0
9
    speech_sample_rate=16000  # 8000, 16000, or 24000 Hz
10
)