MultiagentPersonalAssistant

.idea/MultiagentPersonalAssistant.iml

@@ -2,7 +2,7 @@
 <module type="PYTHON_MODULE" version="4">
   <component name="NewModuleRootManager">
     <content url="file://$MODULE_DIR$" />
-    <orderEntry type="jdk" jdkName="test" jdkType="Python SDK" />
+    <orderEntry type="jdk" jdkName="MultiagentPersonalAssistant" jdkType="Python SDK" />
     <orderEntry type="sourceFolder" forTests="false" />
   </component>
   <component name="PyDocumentationSettings">

.idea/misc.xml

@@ -3,5 +3,5 @@
   <component name="Black">
     <option name="sdkName" value="MultiagentPersonalAssistant" />
   </component>
-  <component name="ProjectRootManager" version="2" project-jdk-name="test" project-jdk-type="Python SDK" />
+  <component name="ProjectRootManager" version="2" project-jdk-name="MultiagentPersonalAssistant" project-jdk-type="Python SDK" />
 </project>

agent/file_preprocessing.py

@@ -1,19 +1,12 @@
-from speechbrain.inference.classifiers import EncoderClassifier
 from langchain_core.messages import HumanMessage
-import speech_recognition as sr
-from pydub import AudioSegment
 from dotenv import load_dotenv
 from PyPDF2 import PdfReader
 from docx import Document
-import torchaudio
 import mimetypes
 import asyncio
 import base64
-import os
 
 load_dotenv()
-language_id = EncoderClassifier.from_hparams(source="speechbrain/lang-id-voxlingua107-ecapa", savedir="tmp")
-
 
 async def preprocess_file(file_name: str):
     mime_type = mimetypes.guess_type(file_name)[0]
@@ -22,49 +15,18 @@ async def preprocess_file(file_name: str):
     elif "video" in mime_type:
         prompt = "Give a detailed description of the video."
     elif "audio" in mime_type:
-        return await asyncio.to_thread(preprocess_audio, file_name)
+        return await preprocess_audio(file_name)
     else:
         return await asyncio.to_thread(preprocess_text, file_name, mime_type)
 
 
-def preprocess_audio(file_name: str):
-    if not os.path.exists(file_name):
-        raise FileNotFoundError(f"File not found: {file_name}")
-
-    wav_file = os.path.splitext(file_name)[0] + ".wav"
-    audio = AudioSegment.from_file(file_name)
-    audio.export(wav_file, format="wav")
-    signal = language_id.load_audio(wav_file)
-    out = language_id.classify_batch(signal)[0].tolist()[0]
-    lang_mapping = {
-        20: "en",
-        106: "zh",
-        35: "hi",
-        22: "es",
-        3: "ar",
-        28: "fr",
-        77: "ru",
-        75: "pt",
-        9: "bn",
-        45: "ja",
-        18: "de",
-        51: "ko",
-        102: "vi",
-        99: "uk"
-    }
-    scores = [out[index] for index in lang_mapping.keys()]
-    language = list(lang_mapping.values())[scores.index(max(scores))]
-    recognizer = sr.Recognizer()
-    with sr.AudioFile(wav_file) as source:
-        audio_data = recognizer.record(source)
-        try:
-            text = recognizer.recognize_google(audio_data, language=language)
-        except sr.UnknownValueError:
-            text = "[Unintelligible audio]"
-        except sr.RequestError as e:
-            text = f"[API error: {e}]"
-    os.remove(wav_file)
-    return text
+async def preprocess_audio(file_name):
+    from agent.models import groq_client
+    transcription = await groq_client.audio.transcriptions.create(
+        model="whisper-large-v3-turbo",
+        file=open(file_name, "rb")
+    )
+    return transcription.text
 
 
 async def preprocess_image(file_name: str):

agent/models.py

@@ -1,6 +1,13 @@
 from langchain.chat_models import init_chat_model
 from dotenv import load_dotenv
+from openai import AsyncOpenAI
+import os
+
 load_dotenv()
+groq_client = AsyncOpenAI(
+    base_url="https://api.groq.com/openai/v1",
+    api_key=os.getenv("GROQ_API_KEY"),
+)
 
 llm_supervisor = init_chat_model(
     model="groq:openai/gpt-oss-120b",
@@ -17,4 +24,4 @@ llm_agents = init_chat_model(
 
 llm_image = init_chat_model(
     model="groq:meta-llama/llama-4-scout-17b-16e-instruct"
-)
+)

requirements.txt

@@ -7,14 +7,13 @@ openai>=1.99.1
 transformers>=4.53.2
 timm>=1.0.19
 langchain-groq>=0.3.7
-torchaudio>=2.8.0
-torch>=2.8.0
 langgraph-supervisor>=0.0.29
 wikipedia>=1.4.0
 wolframalpha>=5.1.3
 python-multipart
 soundfile>=0.13.1
 accelerate
+jsonpointer
 langmem>=0.0.29
 greenlet>=3.2.3
 deepagents>=0.0.3
@@ -27,7 +26,6 @@ pillow>=11.3.0
 langchain-huggingface>=0.3.0
 sentence-transformers>=5.0.0
 pyowm>=3.3.0
-speechbrain>=1.0.3
 langchain-core
 geopy>=2.4.1
 langchain-community>=0.3.27

tmp/classifier.ckpt

@@ -1 +0,0 @@
-/Users/magomedpatahov/.cache/huggingface/hub/models--speechbrain--lang-id-voxlingua107-ecapa/snapshots/0253049ae131d6a4be1c4f0d8b0ff483a0f8c8e9/classifier.ckpt

tmp/embedding_model.ckpt

@@ -1 +0,0 @@
-/Users/magomedpatahov/.cache/huggingface/hub/models--speechbrain--lang-id-voxlingua107-ecapa/snapshots/0253049ae131d6a4be1c4f0d8b0ff483a0f8c8e9/embedding_model.ckpt

tmp/hyperparams.yaml

@@ -1 +0,0 @@
-/Users/magomedpatahov/.cache/huggingface/hub/models--speechbrain--lang-id-voxlingua107-ecapa/snapshots/0253049ae131d6a4be1c4f0d8b0ff483a0f8c8e9/hyperparams.yaml

tmp/label_encoder.ckpt

@@ -1 +0,0 @@
-/Users/magomedpatahov/.cache/huggingface/hub/models--speechbrain--lang-id-voxlingua107-ecapa/snapshots/0253049ae131d6a4be1c4f0d8b0ff483a0f8c8e9/label_encoder.txt