Revert "Clean initial commit for HuggingFace"

This reverts commit f108b29491d90326f90f71fa07edef74d570194b.

.gitignore

@@ -67,4 +67,7 @@ src/database/cvs/tests/*.txt
 .lgcache/
 .langgraph_api/
 
-.idea/
+.idea/
+
+# any .wav files
+*.wav

Dockerfile

@@ -2,8 +2,8 @@ FROM python:3.12-slim
 
 WORKDIR /app
 
-# System dependencies
-RUN apt-get update && apt-get install -y gcc libpq-dev && rm -rf /var/lib/apt/lists/*
+# System dependencies (include Postgres server so DB can run in-container)
+RUN apt-get update && apt-get install -y gcc libpq-dev postgresql postgresql-contrib gosu && rm -rf /var/lib/apt/lists/*
 
 # Copy requirement files
 COPY requirements/base.txt requirements/base.txt
@@ -34,28 +34,8 @@ COPY secrets/ /app/secrets/
 ENV PYTHONPATH=/app
 EXPOSE 7860
 
-# Create entry script inside the image (avoids missing file in build context)
-RUN printf '%s\n' \
-  '#!/usr/bin/env bash' \
-  'set -e' \
-  '' \
-  '# Hugging Face provides PORT; default to 7860 locally' \
-  'export PORT=\"${PORT:-7860}\"' \
-  '' \
-  '# Defaults for local in-container routing; can be overridden via env' \
-  'export SUPERVISOR_API_URL=\"${SUPERVISOR_API_URL:-http://127.0.0.1:8080/api/v1/supervisor}\"' \
-  'export DATABASE_API_URL=\"${DATABASE_API_URL:-http://127.0.0.1:8080/api/v1/db}\"' \
-  'export CV_UPLOAD_API_URL=\"${CV_UPLOAD_API_URL:-http://127.0.0.1:8080/api/v1/cv}\"' \
-  '' \
-  '# Start FastAPI backend' \
-  'uvicorn src.api.app:app --host 0.0.0.0 --port 8080 &' \
-  '' \
-  '# Give the API a moment to come up' \
-  'sleep 2' \
-  '' \
-  '# Run Gradio frontend' \
-  'python src/frontend/gradio/app.py' \
-  > /app/start.sh \
-  && chmod +x /app/start.sh
+# Copy entry script (includes Postgres startup)
+COPY start.sh /app/start.sh
+RUN chmod +x /app/start.sh
 
 CMD ["/app/start.sh"]

README.md

@@ -171,7 +171,7 @@ docker compose --env-file .env -f docker/docker-compose.yml up --build
 The platform orchestrates a complete recruitment pipeline, interacting with both Candidates and the HR Supervisor.
 
 ### 1. The Recruitment Lifecycle
-The system tracks candidates through a defined state machine (see `src/state/candidate.py` for the `CandidateStatus` enum).
+The system tracks candidates through a defined state machine (see `src/backend/state/candidate.py` for the `CandidateStatus` enum).
 
 ```mermaid
 graph TD
@@ -231,7 +231,7 @@ graph TD
 
 To improve the reliability of complex evaluations (such as CV scoring and Voice Interview judging), we enforce **Chain-of-Thought (CoT)** reasoning within our structured outputs, inspired by [Wei et al. (2022)](https://arxiv.org/abs/2201.11903).
 
-By requiring the model to generate a textual explanation *before* assigning numerical scores, we ensure the model "thinks" through the evidence before committing to a decision. This is implemented directly in our Pydantic schemas (e.g., `src/agents/cv_screening/schemas/output_schema.py`), where field order matters:
+By requiring the model to generate a textual explanation *before* assigning numerical scores, we ensure the model "thinks" through the evidence before committing to a decision. This is implemented directly in our Pydantic schemas (e.g., `src/backend/agents/cv_screening/schemas/output_schema.py`), where field order matters:
 
 ```mermaid
 flowchart LR
@@ -362,14 +362,14 @@ A breakdown of the various LLMs, Agents, and Workflows powering the system.
 
 | Component | Type | Model | Description | Location |
 | :--- | :--- | :--- | :--- | :--- |
-| **Supervisor Agent** | 🤖 **Agent** | `gpt-4o` | Orchestrates delegation, planning, and context management. | `src/agents/supervisor/supervisor_v2.py` |
-| **Gmail Agent** | 🤖 **Agent** | `gpt-4o` | Autonomous email management via MCP (read/send/label). | `src/agents/gmail/gmail_agent.py` |
-| **GCalendar Agent** | 🤖 **Agent** | `gpt-4o` | Autonomous calendar scheduling via MCP. | `src/agents/gcalendar/gcalendar_agent.py` |
-| **DB Executor** | 🤖 **Agent** | `gpt-4o` | Writes SQL/Python to query the database (CodeAct). | `src/agents/db_executor/db_executor.py` |
-| **CV Screening** | ⚙️ **Workflow** | `gpt-4o` | Deterministic pipeline: Fetch → Read → Evaluate → Save. | `src/agents/cv_screening/cv_screening_workflow.py` |
-| **Voice Judge** | 🧠 **Simple LLM** | `gpt-4o-audio` | Evaluates audio/transcripts for sentiment & confidence. | `src/agents/voice_screening/judge.py` |
-| **Doc Parser** | 🧠 **Simple LLM** | `gpt-4o-mini` | Vision-based PDF-to-Markdown conversion. | `src/doc_parser/pdf_to_markdown.py` |
-| **History Manager** | 🧠 **Simple LLM** | `gpt-4o-mini` | Summarizes conversation history for context compaction. | `src/context_eng/history_manager.py` |
+| **Supervisor Agent** | 🤖 **Agent** | `gpt-4o` | Orchestrates delegation, planning, and context management. | `src/backend/agents/supervisor/supervisor_v2.py` |
+| **Gmail Agent** | 🤖 **Agent** | `gpt-4o` | Autonomous email management via MCP (read/send/label). | `src/backend/agents/gmail/gmail_agent.py` |
+| **GCalendar Agent** | 🤖 **Agent** | `gpt-4o` | Autonomous calendar scheduling via MCP. | `src/backend/agents/gcalendar/gcalendar_agent.py` |
+| **DB Executor** | 🤖 **Agent** | `gpt-4o` | Writes SQL/Python to query the database (CodeAct). | `src/backend/agents/db_executor/db_executor.py` |
+| **CV Screening** | ⚙️ **Workflow** | `gpt-4o` | Deterministic pipeline: Fetch → Read → Evaluate → Save. | `src/backend/agents/cv_screening/cv_screening_workflow.py` |
+| **Voice Judge** | 🧠 **Simple LLM** | `gpt-4o-audio` | Evaluates audio/transcripts for sentiment & confidence. | `src/backend/agents/voice_screening/judge.py` |
+| **Doc Parser** | 🧠 **Simple LLM** | `gpt-4o-mini` | Vision-based PDF-to-Markdown conversion. | `src/backend/doc_parser/pdf_to_markdown.py` |
+| **History Manager** | 🧠 **Simple LLM** | `gpt-4o-mini` | Summarizes conversation history for context compaction. | `src/backend/context_eng/history_manager.py` |
 
 ### 🔌 ***`Integrated MCP Servers`***
 The system integrates **Model Context Protocol (MCP)** servers to securely and standardizedly connect agents to external tools.

docker/Dockerfile.candidates_db_init

@@ -15,8 +15,10 @@ COPY ../requirements/base.txt ./requirements/base.txt
 COPY ../requirements/db.txt  ./requirements/db.txt
 RUN pip install --no-cache-dir -r requirements/db.txt
 
-# Copy *only* the candidate database module
-COPY src/database/candidates ./src/database/candidates
+# Copy required source modules
+COPY src/backend/database/candidates ./src/backend/database/candidates
+COPY src/backend/state ./src/backend/state
+COPY src/backend/configs ./src/backend/configs
 
 # Default command - use dedicated init script to avoid circular import
-CMD ["python", "-m", "src.database.candidates.init_db"]
+CMD ["python", "-m", "src.backend.database.candidates.init_db"]

docker/Dockerfile.supervisor_api

@@ -39,5 +39,5 @@ COPY .env /app/.env
 EXPOSE 8080
 
 # Run FastAPI with uvicorn
-CMD ["uvicorn", "src.api.app:app", "--host", "0.0.0.0", "--port", "8080"]
+CMD ["uvicorn", "src.backend.api.app:app", "--host", "0.0.0.0", "--port", "8080"]
 

docker/docker-compose.yml

@@ -19,6 +19,10 @@ services:
       interval: 3s
       timeout: 3s
       retries: 5
+    # Hey compose here is env file,
+    # pass it to container, but not the .env itself
+    env_file:
+      - ../.env
     environment:
       POSTGRES_HOST: ${POSTGRES_HOST}
       POSTGRES_PORT: ${POSTGRES_PORT}
@@ -34,18 +38,23 @@ services:
     # Initializes the database or starts the API (depending on command).
     container_name: candidates_db_init
     build:
-      context: .. # build from the project root
+      context: ..  # build from the project root
       dockerfile: docker/Dockerfile.candidates_db_init
     depends_on:
       db:
         condition: service_healthy
+    # Hey compose here is env file, 
+    # pass it to container, but not the .env itself
+    env_file:
+      - ../.env
     environment:
-      POSTGRES_HOST: ${POSTGRES_HOST}
-      POSTGRES_PORT: ${POSTGRES_PORT}
+      # Explicitly set POSTGRES_HOST to the service name 'db' for Docker networking
+      POSTGRES_HOST: db
+      POSTGRES_PORT: 5432
       POSTGRES_USER: ${POSTGRES_USER}
       POSTGRES_PASSWORD: ${POSTGRES_PASSWORD}
       POSTGRES_DB: ${POSTGRES_DB}
-    # command: ["python", "-m", "src.database.candidates.init_db"]
+    command: ["python", "-m", "src.backend.database.candidates.init_db"]
 
     volumes:
       # --- Local code mount (for development only) ---
@@ -53,7 +62,7 @@ services:
       # into the container at /app.
       # ✅ Enables live code changes without rebuilding the image.
       # ⚠️ Do NOT use in production – overrides the built image code.
-      - ../:/app # optional: live reload for local dev
+      - ../:/app  # optional: live reload for local dev
 
     networks:
       - hrnet
@@ -69,15 +78,17 @@ services:
     depends_on:
       - db
       - supervisor_api
+    env_file:
+      - ../.env
     environment:
       # Database connection
-      POSTGRES_HOST: ${POSTGRES_HOST}
-      POSTGRES_PORT: ${POSTGRES_PORT}
+      POSTGRES_HOST: db
+      POSTGRES_PORT: 5432
       POSTGRES_USER: ${POSTGRES_USER}
       POSTGRES_PASSWORD: ${POSTGRES_PASSWORD}
       POSTGRES_DB: ${POSTGRES_DB}
-      DATABASE_URL: postgresql://${POSTGRES_USER}:${POSTGRES_PASSWORD}@${POSTGRES_HOST}:${POSTGRES_PORT}/${POSTGRES_DB}
-      CV_UPLOAD_PATH: /app/src/database/cvs/uploads
+      DATABASE_URL: postgresql://${POSTGRES_USER}:${POSTGRES_PASSWORD}@db:5432/${POSTGRES_DB}
+      CV_UPLOAD_PATH: /app/src/backend/database/cvs/uploads
       # App specific
       CV_UPLOAD_API_URL: http://supervisor_api:8080/api/v1/cv
       PYTHONPATH: /app
@@ -85,15 +96,8 @@ services:
       # Mount local code for live updates
       - ../:/app
       # Shared volume for CV uploads (persistent)
-      - ../src/database/cvs:/app/src/database/cvs
-    command:
-      [
-        "streamlit",
-        "run",
-        "src/frontend/streamlit/cv_ui/app.py",
-        "--server.port=8501",
-        "--server.address=0.0.0.0",
-      ]
+      - ../src/backend/database/cvs:/app/src/backend/database/cvs
+    command: ["streamlit", "run", "src/frontend/streamlit/cv_ui/app.py", "--server.port=8501", "--server.address=0.0.0.0"]
     networks:
       - hrnet
 
@@ -105,6 +109,8 @@ services:
       dockerfile: docker/Dockerfile.voice_proxy
     ports:
       - "8000:8000"
+    env_file:
+      - ../.env
     depends_on:
       - db
       - candidates_db_init
@@ -112,20 +118,16 @@ services:
       PYTHONPATH: /app
       OPENAI_API_KEY: ${OPENAI_API_KEY}
       BACKEND_API_URL: http://supervisor_api:8080
+      # Database connection
+      POSTGRES_HOST: db
+      POSTGRES_PORT: 5432
+      POSTGRES_USER: ${POSTGRES_USER}
+      POSTGRES_PASSWORD: ${POSTGRES_PASSWORD}
+      POSTGRES_DB: ${POSTGRES_DB}
     volumes:
       # Mount local code for live updates
       - ../:/app
-    command:
-      [
-        "python",
-        "-m",
-        "uvicorn",
-        "src.frontend.streamlit.voice_screening_ui.proxy:app",
-        "--host",
-        "0.0.0.0",
-        "--port",
-        "8000",
-      ]
+    command: ["python", "-m", "uvicorn", "src.frontend.streamlit.voice_screening_ui.proxy:app", "--host", "0.0.0.0", "--port", "8000"]
     networks:
       - hrnet
 
@@ -136,10 +138,12 @@ services:
       context: ..
       dockerfile: docker/Dockerfile.voice_screening
     ports:
-      - "8502:8501" # Map host port 8502 to container port 8501
+      - "8502:8501"  # Map host port 8502 to container port 8501
     depends_on:
       - db
       - websocket_proxy
+    env_file:
+      - ../.env
     environment:
       DATABASE_URL: postgresql://agentic_user:password123@db:5432/agentic_hr
       PYTHONPATH: /app
@@ -148,14 +152,7 @@ services:
     volumes:
       # Mount local code for live updates
       - ../:/app
-    command:
-      [
-        "streamlit",
-        "run",
-        "src/frontend/streamlit/voice_screening_ui/app.py",
-        "--server.port=8501",
-        "--server.address=0.0.0.0",
-      ]
+    command: ["streamlit", "run", "src/frontend/streamlit/voice_screening_ui/app.py", "--server.port=8501", "--server.address=0.0.0.0"]
     networks:
       - hrnet
 
@@ -166,13 +163,15 @@ services:
       context: ..
       dockerfile: docker/Dockerfile.supervisor_api
     ports:
-      - "8080:8080" # Map host port 8080 to container port 8080
+      - "8080:8080"  # Map host port 8080 to container port 8080
     depends_on:
       - db
+    env_file:
+      - ../.env
     environment:
       # We set POSTGRES_HOST to 'db' so the agent connects to the container internal network
-      POSTGRES_HOST: ${POSTGRES_HOST}
-      POSTGRES_PORT: ${POSTGRES_PORT}
+      POSTGRES_HOST: db
+      POSTGRES_PORT: 5432
       POSTGRES_USER: ${POSTGRES_USER}
       POSTGRES_PASSWORD: ${POSTGRES_PASSWORD}
       POSTGRES_DB: ${POSTGRES_DB}
@@ -180,19 +179,12 @@ services:
       PROMPTLAYER_API_KEY: ${PROMPTLAYER_API_KEY}
       OPENAI_API_KEY: ${OPENAI_API_KEY}
       WEBSOCKET_PROXY_URL: ws://websocket_proxy:8000/ws/realtime
+      CV_UPLOAD_PATH: /app/src/backend/database/cvs/uploads
+      CV_PARSED_PATH: /app/src/backend/database/cvs/parsed
     volumes:
       # Mount local code for live updates
       - ../:/app
-    command:
-      [
-        "uvicorn",
-        "src.api.app:app",
-        "--host",
-        "0.0.0.0",
-        "--port",
-        "8080",
-        "--reload",
-      ]
+    command: ["uvicorn", "src.backend.api.app:app", "--host", "0.0.0.0", "--port", "8080", "--reload"]
     networks:
       - hrnet
 
@@ -203,10 +195,12 @@ services:
       context: ..
       dockerfile: docker/Dockerfile.supervisor
     ports:
-      - "8503:8501" # Map host port 8503 to container port 8501
+      - "8503:8501"  # Map host port 8503 to container port 8501
     depends_on:
       - db
       - supervisor_api
+    env_file:
+      - ../.env
     environment:
       # We set POSTGRES_HOST to 'db' so the agent connects to the container internal network
       PYTHONPATH: /app

docker/info.md

@@ -20,6 +20,7 @@
 docker compose --env-file .env -f docker/docker-compose.yml up --build
 ```
 
+
 ---
 
 ### Resetting the Environment
@@ -30,11 +31,14 @@ To completely reset the environment and database:
 
 ```bash
 # 1. Stop running containers
-docker compose -f docker/docker-compose.yml down
+docker compose -f docker/docker-compose.yml down --remove-orphans
 
 # 2. Remove the persistent database volume
 docker volume rm docker_postgres_data
 
-# 3. Rebuild and start fresh
+# 3. Prune unused Docker resources (optional but recommended)
+docker system prune -f
+
+# 4. Rebuild and start fresh
 docker compose --env-file .env -f docker/docker-compose.yml up --build
 ```

scripts/db/list_candidates.py

@@ -10,8 +10,8 @@ from sqlalchemy.exc import ProgrammingError
 # Ensure project root is in path
 import scripts.db  # noqa: F401
 
-from src.database.candidates.client import SessionLocal
-from src.database.candidates.models import Candidate
+from src.backend.database.candidates.client import SessionLocal
+from src.backend.database.candidates.models import Candidate
 
 
 def list_candidates(limit: int = 10) -> bool:
@@ -41,7 +41,17 @@ def list_candidates(limit: int = 10) -> bool:
                 .all()
             )
             for c in candidates:
-                print(f" - {c.full_name} | {c.email} | Status: {c.status}")
+                print(f" - ID: {c.id}")
+                print(f"   Full Name: {c.full_name}")
+                print(f"   Email: {c.email}")
+                print(f"   Phone: {c.phone_number}")
+                print(f"   CV Path: {c.cv_file_path}")
+                print(f"   Parsed CV Path: {c.parsed_cv_file_path}")
+                print(f"   Status: {c.status}")
+                print(f"   Auth Code: {c.auth_code}")
+                print(f"   Created At: {c.created_at}")
+                print(f"   Updated At: {c.updated_at}")
+                print("-" * 40)
         
         return True
 

scripts/db/test_connection.py

@@ -11,7 +11,7 @@ from sqlalchemy import text
 # Ensure project root is in path
 import scripts.db  # noqa: F401
 
-from src.database.candidates.client import get_engine
+from src.backend.database.candidates.client import get_engine
 
 
 def test_connection() -> bool:

scripts/db/test_session.py

@@ -10,7 +10,7 @@ from sqlalchemy import text
 # Ensure project root is in path
 import scripts.db  # noqa: F401
 
-from src.database.candidates.client import SessionLocal
+from src.backend.database.candidates.client import SessionLocal
 
 
 def test_session_query() -> bool:

scripts/db/wipe.py

@@ -11,7 +11,7 @@ from sqlalchemy import text
 project_root = os.path.abspath(os.path.join(os.path.dirname(__file__), '../../'))
 sys.path.append(project_root)
 
-from src.database.candidates.client import get_engine
+from src.backend.database.candidates.client import get_engine
 
 def wipe_database():
     print("⚠️  WARNING: This will PERMANENTLY DELETE ALL RECORDS from the 'candidates' table and all related tables (CASCADE).")

src/agents/db_executor/codeact/core/codeact.py

@@ -319,6 +319,13 @@ class CodeActAgent:
                 serializable_new_vars = self._filter_serializable(new_vars)
                 new_context = {**existing_context, **serializable_new_vars}
 
+                # Format output properly
+                content_str = (
+                    f"Sandbox result of your executed code:\n{json.dumps(output, default=str)}"
+                    if not isinstance(output, str)
+                    else f"Sandbox result of your executed code:\n{output}"
+                )
+
                 # Return OpenAI-compliant tool result
                 return {
                     "messages": [
@@ -326,13 +333,7 @@ class CodeActAgent:
                             "role": "tool",
                             "tool_call_id": tool_call_id,
                             "name": "sandbox",
-                            "content": (
-                                f"Sandbox result of your executed code:\n{json.dumps(output)}"
-                                if not isinstance(output, str)
-                                else f"Sandbox result of your executed code:\n{output}"
-                                # Keep as string if already string else JSON serialize
-                            ),
-
+                            "content": content_str
                         }
                     ],
                     "context": new_context,
@@ -363,6 +364,13 @@ class CodeActAgent:
                 serializable_new_vars = self._filter_serializable(new_vars)
                 new_context = {**existing_context, **serializable_new_vars}
 
+                # Format output properly
+                content_str = (                    # NOTE: before "json.dumps(output)"
+                    f"Sandbox result of your executed code:\n{json.dumps(output, default=str)}"
+                    if not isinstance(output, str)
+                    else f"Sandbox result of your executed code:\n{output}"
+                )
+
                 # Return OpenAI-compliant tool result
                 return {
                     "messages": [
@@ -370,11 +378,7 @@ class CodeActAgent:
                             "role": "tool",
                             "tool_call_id": tool_call_id,
                             "name": "sandbox",
-                            "content": (
-                                f"Sandbox result of your executed code:\n{json.dumps(output)}"
-                                if not isinstance(output, str)
-                                else f"Sandbox result of your executed code:\n{output}"
-                            ),
+                            "content": content_str,
                             # Keep as string if already string else JSON serialize      
                         }
                     ],
@@ -538,4 +542,3 @@ if __name__ == "__main__":
                     pretty_print_state(chunk, show_context=False)
 
         print("\n")
-

src/agents/db_executor/db_executor.py

@@ -11,6 +11,7 @@ from langchain_core.tools import tool
 from typing import Dict, Any
 from src.prompts import get_prompt
 from src.database.candidates import evaluate_cv_screening_decision
+from src.state.candidate import CandidateStatus, InterviewStatus, DecisionStatus
 
 
 SYSTEM_PROMPT = get_prompt(
@@ -40,6 +41,9 @@ def db_executor(query: str) -> str:
         "VoiceScreeningResult": VoiceScreeningResult,
         "InterviewScheduling": InterviewScheduling,
         "FinalDecision": FinalDecision,
+        "CandidateStatus": CandidateStatus,
+        "InterviewStatus": InterviewStatus,
+        "DecisionStatus": DecisionStatus,
     }
 
     try:
@@ -77,19 +81,3 @@ def db_executor(query: str) -> str:
 
 
 
-if __name__ == "__main__":
-    from rich.console import Console
-    from rich.panel import Panel
-
-    console = Console()
-    query = "Fetch all candidates and their status."
-    
-    console.rule("[bold magenta]DB Executor Test Run[/bold magenta]")
-    console.print(f"[cyan]Query:[/] {query}\n")
-
-    result = db_executor(query)
-
-    # 🧠 Show model result nicely
-    console.print(Panel.fit(result, title="🧠 Model Output", border_style="blue"))
-
-    console.rule("[bold green]End of Execution[/bold green]")

src/agents/gcalendar/gcalendar_agent.py

@@ -34,8 +34,7 @@ def gcalendar_agent(query: str) -> str:
             # Load settings
             settings = GoogleCalendarSettings()
             CALENDAR_MCP_DIR = settings.calendar_mcp_dir
-            CREDS = settings.creds
-            TOKEN = settings.token
+            CREDS, TOKEN = settings.materialize_files()
 
             # Initialize model
             model = ChatOpenAI(model="gpt-4o", temperature=0)

src/agents/gmail/gmail_agent.py

@@ -1,6 +1,8 @@
 import asyncio
+import os
 import shutil
 from pathlib import Path
+from pydantic_core import ValidationError
 from langchain_core.tools import tool
 from langchain_mcp_adapters.client import MultiServerMCPClient
 from langchain.agents import create_agent
@@ -39,11 +41,24 @@ def gmail_agent(query: str) -> str:
     if not UV_PATH:
         return "❌ Error: 'uv' executable not found. Please ensure uv is installed and in the system PATH."
 
+    # Validate required Gmail settings before proceeding
+    creds_env = os.getenv("GMAIL_CREDS_JSON")
+    token_env = os.getenv("GMAIL_TOKEN_JSON")
+    if not creds_env or not token_env:
+        return "❌ Gmail credentials not configured. Set GMAIL_CREDS_JSON and GMAIL_TOKEN_JSON environment variables."
+
     try:
         import asyncio
         async def _run_async():
             # Load settings
-            settings = GMailSettings()
+            try:
+                # Pass env values directly to avoid reliance on env file loading
+                settings = GMailSettings(creds_json=creds_env, token_json=token_env)
+            except ValidationError as ve:
+                return f"❌ Gmail settings invalid: {ve}"
+            except Exception as e:
+                return f"❌ Gmail settings error: {e}"
+            creds_path, token_path = settings.materialize_files()
             
             # Initialize model
             model = ChatOpenAI(model="gpt-4o", temperature=0)
@@ -56,8 +71,8 @@ def gmail_agent(query: str) -> str:
                         "args": [
                             "--directory", str(settings.gmail_mcp_dir),
                             "run", "gmail",
-                            "--creds-file-path", str(settings.creds),
-                            "--token-path", str(settings.token),
+                            "--creds-file-path", str(creds_path),
+                            "--token-path", str(token_path),
                         ],
                         "transport": "stdio",
                     }

src/context_eng/compacting_supervisor.py

@@ -135,7 +135,7 @@ history_manager = HistoryManager(memory_saver=memory)
 compacting_supervisor = CompactingSupervisor(
     agent=supervisor_agent, 
     history_manager=history_manager,
-    token_limit=500,
+    token_limit=2500,
     compaction_ratio=0.5
 )
 

src/database/candidates/client.py

@@ -32,4 +32,4 @@ def get_engine():
 
 # --- SQLAlchemy session setup ---
 engine = get_engine()
-SessionLocal = sessionmaker(bind=engine, autoflush=False, autocommit=False)
+SessionLocal = sessionmaker(bind=engine, autoflush=False, autocommit=False)

src/database/candidates/init_db.py

@@ -11,7 +11,7 @@ Usage:
 
 from src.database.candidates.client import engine
 from src.database.candidates.models import Base
-
+from sqlalchemy import inspect
 
 def init_db():
     """
@@ -19,13 +19,24 @@ def init_db():
     Intended for dev setup / Docker initialization.
     """
     try:
+        print("🚀 Starting database initialization...")
         Base.metadata.create_all(bind=engine)
-        print("✅ Database initialized successfully.")
+        
+        # Verify tables
+        inspector = inspect(engine)
+        tables = inspector.get_table_names()
+        print(f"📊 Found tables: {tables}")
+        
+        if "candidates" in tables:
+            print("✅ Database initialized successfully.")
+            return True
+        else:
+            print("❌ Error: 'candidates' table was not created!")
+            
     except Exception as e:
         print(f"❌ Failed to initialize database: {e}")
         raise
 
 
 if __name__ == "__main__":
-    init_db()
-
+    init_db()

src/database/candidates/models.py

@@ -22,8 +22,11 @@ Base = declarative_base()
 
 
 def generate_auth_code() -> str:
-    """Generate a 6-digit random authentication code."""
-    return "".join(secrets.choice(string.digits) for _ in range(6))
+    """Generate a 6-digit random authentication code.
+    """
+    return "".join(
+        secrets.choice(string.digits) for _ in range(6)
+    )
 
 # --- TABLES ---
 
@@ -126,4 +129,4 @@ class FinalDecision(Base):
     human_notes = Column(Text)
     timestamp = Column(DateTime, default=datetime.utcnow)
 
-    candidate = relationship("Candidate", back_populates="final_decision")
+    candidate = relationship("Candidate", back_populates="final_decision")

src/database/context/__init__.py

@@ -1,153 +0,0 @@
-from sqlalchemy import (
-    Column, String, Integer, Float, Enum, DateTime, Text, ForeignKey, JSON
-)
-from sqlalchemy.dialects.postgresql import UUID
-from sqlalchemy.orm import declarative_base, relationship
-from datetime import datetime
-import enum
-import uuid
-
-Base = declarative_base()
-
-
-# ==============================================================
-# ENUM DEFINITIONS
-# ==============================================================
-
-class CandidateStatus(enum.Enum):
-    APPLIED = "applied"
-    CV_SCREENED = "cv_screened"
-    INVITED_VOICE = "invited_voice"
-    VOICE_DONE = "voice_done"
-    SCHEDULED_HR = "scheduled_hr"
-    DECISION_PENDING = "decision_pending"
-    REJECTED = "rejected"
-    HIRED = "hired"
-
-
-class InterviewStatus(enum.Enum):
-    SCHEDULED = "scheduled"
-    COMPLETED = "completed"
-    CANCELLED = "cancelled"
-
-
-class Decision(enum.Enum):
-    HIRE = "hire"
-    REJECT = "reject"
-    MAYBE = "maybe"
-
-
-# ==============================================================
-# MAIN TABLES
-# ==============================================================
-
-class Candidate(Base):
-    __tablename__ = "candidates"
-
-    id = Column(UUID(as_uuid=True), primary_key=True, default=uuid.uuid4)
-    full_name = Column(String(255), nullable=False)
-    email = Column(String(255), nullable=False, unique=True)
-    phone_number = Column(String(50), nullable=True)
-    cv_file_path = Column(String(500), nullable=True)
-    parsed_cv_json = Column(JSON, nullable=True)
-    status = Column(Enum(CandidateStatus), default=CandidateStatus.APPLIED)
-    created_at = Column(DateTime, default=datetime.utcnow)
-    updated_at = Column(DateTime, default=datetime.utcnow, onupdate=datetime.utcnow)
-
-    # Relationships
-    cv_results = relationship("CVScreeningResult", back_populates="candidate", cascade="all, delete-orphan")
-    voice_results = relationship("VoiceScreeningResult", back_populates="candidate", cascade="all, delete-orphan")
-    interviews = relationship("InterviewScheduling", back_populates="candidate", cascade="all, delete-orphan")
-    decision = relationship("FinalDecision", back_populates="candidate", uselist=False, cascade="all, delete-orphan")
-
-
-# ==============================================================
-# CV SCREENING RESULTS
-# ==============================================================
-
-class CVScreeningResult(Base):
-    __tablename__ = "cv_screening_results"
-
-    id = Column(UUID(as_uuid=True), primary_key=True, default=uuid.uuid4)
-    candidate_id = Column(UUID(as_uuid=True), ForeignKey("candidates.id"), nullable=False)
-    job_title = Column(String(255), nullable=True)
-
-    skills_match_score = Column(Float, nullable=True)
-    experience_match_score = Column(Float, nullable=True)
-    education_match_score = Column(Float, nullable=True)
-    overall_fit_score = Column(Float, nullable=True)
-
-    llm_feedback = Column(Text, nullable=True)
-    reasoning_trace = Column(JSON, nullable=True)
-
-    timestamp = Column(DateTime, default=datetime.utcnow)
-
-    candidate = relationship("Candidate", back_populates="cv_results")
-
-
-# ==============================================================
-# VOICE SCREENING RESULTS
-# ==============================================================
-
-class VoiceScreeningResult(Base):
-    __tablename__ = "voice_screening_results"
-
-    id = Column(UUID(as_uuid=True), primary_key=True, default=uuid.uuid4)
-    candidate_id = Column(UUID(as_uuid=True), ForeignKey("candidates.id"), nullable=False)
-
-    call_sid = Column(String(255), nullable=True)
-    transcript_text = Column(Text, nullable=True)
-
-    sentiment_score = Column(Float, nullable=True)
-    confidence_score = Column(Float, nullable=True)
-    communication_score = Column(Float, nullable=True)
-
-    llm_summary = Column(Text, nullable=True)
-    llm_judgment_json = Column(JSON, nullable=True)
-    audio_url = Column(String(500), nullable=True)
-
-    timestamp = Column(DateTime, default=datetime.utcnow)
-
-    candidate = relationship("Candidate", back_populates="voice_results")
-
-
-# ==============================================================
-# INTERVIEW SCHEDULING
-# ==============================================================
-
-class InterviewScheduling(Base):
-    __tablename__ = "interview_scheduling"
-
-    id = Column(UUID(as_uuid=True), primary_key=True, default=uuid.uuid4)
-    candidate_id = Column(UUID(as_uuid=True), ForeignKey("candidates.id"), nullable=False)
-
-    calendar_event_id = Column(String(255), nullable=True)
-    event_summary = Column(String(255), nullable=True)
-
-    start_time = Column(DateTime, nullable=True)
-    end_time = Column(DateTime, nullable=True)
-    status = Column(Enum(InterviewStatus), default=InterviewStatus.SCHEDULED)
-
-    timestamp = Column(DateTime, default=datetime.utcnow)
-
-    candidate = relationship("Candidate", back_populates="interviews")
-
-
-# ==============================================================
-# FINAL DECISION
-# ==============================================================
-
-class FinalDecision(Base):
-    __tablename__ = "final_decision"
-
-    id = Column(UUID(as_uuid=True), primary_key=True, default=uuid.uuid4)
-    candidate_id = Column(UUID(as_uuid=True), ForeignKey("candidates.id"), nullable=False, unique=True)
-
-    overall_score = Column(Float, nullable=True)
-    decision = Column(Enum(Decision), default=Decision.MAYBE)
-    llm_rationale = Column(Text, nullable=True)
-    human_notes = Column(Text, nullable=True)
-
-    timestamp = Column(DateTime, default=datetime.utcnow)
-
-    candidate = relationship("Candidate", back_populates="decision")

src/frontend/streamlit/voice_screening_ui/proxy.py

@@ -22,8 +22,8 @@ from dotenv import load_dotenv
 from sqlalchemy import select
 
 # Import database client and models
-from src.database.candidates.client import SessionLocal
-from src.database.candidates.models import Candidate
+from src.backend.database.candidates.client import SessionLocal
+from src.backend.database.candidates.models import Candidate
 
 load_dotenv()
 

src/prompts/templates/cv_screener/v1.txt

@@ -1,10 +1,8 @@
 You are an HR assistant evaluating how well a candidate's CV matches a given 
 job description. Generate a concise assessment summary first to ground your 
 reasoning. Then assign calibrated match scores between 0 and 1.
-
 The scores should be based on the following criteria:
     1. Skills Match Score: How well the candidate's skills match the job description."
     2. Experience Match Score: How well the candidate's experience matches the job description.
     3. Education Match Score: How well the candidate's education matches the job description.
     4. Overall Fit Score: How well the candidate's CV fits the job description.
-

src/prompts/templates/db_executor/v1.txt

@@ -1,56 +1,46 @@
-You are the **Database Executor Agent**, responsible for generating 
-and executing **SQLAlchemy ORM-style** Python code on behalf of the HR Supervisor Agent.
-
-Your job: perform safe and deterministic **read/write/update operations** 
-in the HR recruitment database, based on clear natural-language requests.
-
+You are the **Database Executor Agent**, responsible for generating and executing **SQLAlchemy ORM-style** Python code on behalf of the HR Supervisor Agent.
+Your job: perform safe and deterministic **read/write/update operations** in the HR recruitment database, based on clear natural-language requests.
 ---
-
-### ✅ Rules
-1. Use SQLAlchemy ORM — not raw SQL.
-2. Use `session` (provided) for all queries.
-3. Return clean Python dict or list results — no ORM objects.
+# Rules
+1. Use SQLAlchemy ORM not raw SQL.
+2. **MANDATORY**: Use the pre-provided `session` object from the global context.
+   - ❌ NEVER do `create_engine` or `sessionmaker`.
+   - ❌ NEVER import `sqlite3` or `psycopg2` directly.
+   - ✅ ALWAYS use `session.query(...)`.
+3. Return clean Python dict or list results, no ORM objects.
 4. Commit only when needed (`session.commit()`).
 5. Never alter schema, connection, or delete/drop tables.
 6. Validate record existence before updating or inserting.
 7. Briefly explain what was done in plain English.
-
 ---
-
-### 🧩 Database Overview (ORM Models)
-**Note**: All these models are already imported and available in the global context. 
-**DO NOT** try to import them again. Use them directly (e.g. `session.query(Candidate)...`).
-
+# Database Overview (ORM Models)
+**Note**: All these models AND Enums (`CandidateStatus`, `InterviewStatus`, `DecisionStatus`) are already imported and available in the global context.
+**DO NOT** try to import them again. Use them directly (e.g. `session.query(Candidate)...` or `status=CandidateStatus.hired`).
+**DATABASE TYPE**: PostgreSQL (managed by the system). DO NOT assume SQLite.
 **Candidate**
 - id (UUID, PK)  
 - full_name, email (unique), phone_number  
-- cv_file_path, parsed_cv_file_path  
-- status (Enum: `applied`, `cv_screened`, `cv_passed`, `cv_rejected`, `voice_passed`, `voice_rejected`, `interview_scheduled`, `decision_made`)  
+- cv_file_path, parsed_cv_file_path, created_at, updated_at, auth_code  
+- status (Enum `CandidateStatus`: `applied`, `cv_screened`, `cv_passed`, `cv_rejected`, `voice_invitation_sent`, `voice_done`, `voice_passed`, `voice_rejected`, `interview_scheduled`, `interview_passed`, `interview_rejected`, `decision_made`, `hired`, `rejected`)
 - Relationships → `cv_screening_results`, `voice_screening_results`, `interview_scheduling`, `final_decision`
-
 **CVScreeningResult**
 - candidate_id → Candidate.id  
-- skills_match_score, experience_match_score, education_match_score, overall_fit_score  
+- job_title, skills_match_score, experience_match_score, education_match_score, overall_fit_score  
 - llm_feedback, reasoning_trace (JSON), timestamp  
-
 **VoiceScreeningResult**
 - candidate_id → Candidate.id  
-- transcript_text, sentiment_score, communication_score, confidence_score  
+- call_sid, transcript_text, sentiment_score, communication_score, confidence_score  
 - llm_summary, llm_judgment_json, audio_url, timestamp  
-
 **InterviewScheduling**
 - candidate_id → Candidate.id  
-- calendar_event_id, start_time, end_time  
-- status (Enum: `scheduled`, `completed`, `cancelled`)
-
+- calendar_event_id, event_summary, start_time, end_time  
+- status (Enum `InterviewStatus`: `scheduled`, `completed`, `cancelled`, `passed`, `rejected`)
 **FinalDecision**
 - candidate_id → Candidate.id  
-- overall_score, decision (Enum: `hire`, `reject`, `maybe`)  
+- overall_score, decision (Enum `DecisionStatus`: `hired`, `rejected`, `pending`)  
 - llm_rationale, human_notes, timestamp  
-
 ---
-
-🧾 Expected Execution Pattern
+Expected Execution Pattern
 When asked to perform a task, you must:
 1. Construct ORM-based Python code using session and the given models.
 2. Store final results in a variable named result.
@@ -60,15 +50,10 @@ import json
 print(json.dumps(result, indent=2, default=str))
 ```
 4. Optionally, include a short explanatory comment after the code.
-
-### 🧾 Output Format
+# Output Format
 1. **Execution:** Your Python code must `print()` the results so they are visible in the tool output.
-2. **Final Response:** After the code runs, provide a **clear, natural language summary** of what you found or did.
-   - *Example:* "I successfully updated the status for Sebastian Wefers to 'scheduled'."
-   - *Example:* "I retrieved 3 candidates: John, Jane, and Bob."
-
-### 🚨 Error Handling
+2. **Final Response:** After the code runs, provide a **clear, natural language summary** of what you found or did. It should be clear enough that a random person would understand.
+# Error Handling
 If you encounter errors:
 1. **Self-Correction:** Attempt to fix the code and retry within the reasoning loop.
-2. **Terminal Failure:** If you cannot resolve the issue, explain the problem clearly to the user in plain English.
-   - *Example:* "I tried to update the record, but I could not find a candidate with that email address."
+2. **Terminal Failure:** If you cannot resolve the issue, explain the problem clearly in plain English. Provide verbatim snippets of the error.

src/prompts/templates/gcalendar/v1.txt

@@ -1,10 +1,6 @@
-You are a scheduling assistant authorized to use Google Calendar MCP tools.
-You can for instance list, create, and analyze events. 
-
+You are a scheduling assistant authorized to use Google Calendar MCP tools. You can for instance list, create, and analyze events. 
 IMPORTANT: 
 - For any requests regarding "my calendar", "my availability", or general scheduling without specific attendees, assume the "primary" calendar. 
 - You do NOT need to ask for a calendar ID for the user; the system defaults to their primary calendar.
 - Only ask for calendar IDs if the user asks about a specific third party whose email/ID is not known.
-
-Always confirm the action taken and if an error occurs report it back 
-for transparency and troubleshooting.
+Always confirm the action taken and if an error occurs report it back verbatim.

src/prompts/templates/supervisor/v1.txt

@@ -1,50 +1,64 @@
-You are the **Supervisor Agent** overseeing the entire recruitment workflow.  
-You act on behalf of the HR manager **Casey Jordan** (`hr.cjordan.agent.hack.winter25@gmail.com`), 
-who is the only person talking to you.
-
+You are the **Supervisor Agent** overseeing the entire recruitment workflow. You act on behalf of the HR manager **Casey Jordan** (`hr.cjordan.agent.hack.winter25@gmail.com`), who is the only person talking to you.
 Understand the candidate lifecycle status flow:
-`applied` → `cv_screened`  → `interview_scheduled` → `decision_made`.
-
+1. `applied` (Application received)
+2. `cv_screened` (CV Analyzed)
+3. `cv_passed` or `cv_rejected` (Outcome of CV Screening)
+4. `voice_invitation_sent` (If CV Passed)
+5. `voice_done` (Candidate completed AI Voice Interview)
+6. `voice_passed` or `voice_rejected` (Outcome of Voice Analysis)
+7. `interview_scheduled` (Final Human Interview)
+8. `decision_made` (Final Offer or Rejection)
+9. `hired` (If the decision is positive and confirmed)
+10. `rejected` (If the decision is negative)
 ---
-
-### 🎯 Your Role
+# Your Role
 You coordinate and supervise the hiring process from CV submission to final decision.  
 You have access to specialized sub-agents that handle:
 - Database operations (querying, updating, reporting)
 - CV screening and evaluation
+- Voice screening and analysis
 - Email communication (for candidates and Casey)
 - Calendar scheduling (for HR meetings and interviews)
-
-You do **not** perform these actions yourself — instead, you **delegate** to sub-agents when needed.
+You do **not** perform these actions yourself: instead, you **delegate** to sub-agents when needed.
 ---
-
-### ⚙️ Recruitment Process Overview
+# Recruitment Process Overview
 1. **Application submitted** → Candidate starts with status `applied`.
 2. **CV screening** → 
    - Run `cv_screening_workflow` (updates status to `cv_screened` automatically).
    - Ask `db_executor` to "evaluate screening results" (updates status to `cv_passed` or `cv_rejected`). 
      Here you can optionally specify a minimum passing score (default is 7.0).
-3. **Notification** →  
+3. **Voice Screening Invitation** →  
    - If `cv_rejected`, send a polite rejection email.  
-   - If `cv_passed`, send an email requesting available time slots for a voice or in-person interview.
-4. **Scheduling** → 
-   - Use the calendar agent to check **our (HR)** availability (`primary` calendar).
-   - You CANNOT check the candidate's calendar. You must **ask** the candidate for their preferred times via email.
-   - Once a time is agreed upon, use the calendar agent to schedule the interview.
-5. **Decision** → Once interviews are complete, record and communicate the final decision.
-
+   - If `cv_passed`, fetch `auth_code` from DB and email the candidate the voice screening invitation including this code for login.
+   - Update status to `voice_invitation_sent` via `db_executor`.
+4. **Voice Screening** →
+   - Candidates complete the AI voice interview.
+   - The system updates status to `voice_done` automatically.
+   - Ask `voice_judge` to "evaluate voice screening results" (this automatically updates status to `voice_passed` or `voice_rejected`).
+5. **Interview Invitation (Person-to-Person)** →
+   - If `voice_rejected`, send a polite rejection email.
+   - If `voice_passed`:
+     - Use the calendar agent to check **HR availability** for this and next week (`primary` calendar).
+     - Send a success email to the candidate suggesting these available time slots and asking for their preference.
+6. **Scheduling** → 
+   - Once the candidate replies with a preferred time, use the calendar agent to schedule the interview.
+   - Update status to `interview_scheduled`.
+7. **Final Decision** → 
+   - Once interviews are complete, record the final decision in the database.
+   - Update candidate status to `decision_made`.
+   - Create or update the `FinalDecision` record with the decision (`hired`, `rejected`, or `pending`).
+   - If `hired`, update candidate status to `hired`.
+   - If `rejected`, update candidate status to `rejected`.
+   - Communicate the result to the candidate via email.
 Always notify Casey what a status was updated to.
 ---
-
-### 🧠 Reasoning & Planning Strategy
+# Reasoning & Planning Strategy
 Before calling tools, **THINK**:
 1. **Sequential Dependencies (Action A → Action B):** If Action B requires data (like an email address), perform Action A (fetch data) first.
    - **Example:** Before asking `gmail_agent` to send an email, you **must always** ask `db_executor` to retrieve the candidate's email address first.
-2. **Robust DB Instructions:** ALWAYS ask the `db_executor` to "**Create or update** the record" when changing status. NEVER just ask to "Update", as the record might not exist yet.
-
-
-### 🧠 Your Behavior
+2. **Robust DB Instructions:** ALWAYS ask the `db_executor` to "**create or update** the record" when changing status. NEVER just ask to "Update", as the record might not exist yet.
+# Your Behavior
 - Use the available sub-agents for all database queries, screenings, email sends, and calendar operations.  
-- Respond clearly, professionally and comprehensively to Casey’s requests. 
-- Always share with Casey what actions you have taken and what results were produced.  
-- If you or any sub-agent encounter an error, **notify the Casey immediately** for troubleshooting.
+- Respond clearly, professionally and comprehensively to the user's requests. 
+- Always share with the user what actions you have taken and what results were produced.  
+- If you or any sub-agent encounter an error, **notify the user immediately**.

src/state/candidate.py

@@ -16,9 +16,19 @@ class CandidateStatus(str, enum.Enum):
         -> "voice_invitation_sent"
 
     4) Voice Screening
-        -> "voice_screened"
-              & "voice_passed" 
-              OR "voice_rejected"
+        -> "voice_done"
+        -> "voice_passed" 
+        -> "voice_rejected"
+        
+    5) Interview Scheduling
+        -> "interview_scheduled"
+        -> "interview_passed"
+        -> "interview_rejected"
+
+    6) Final Decision
+        -> "decision_made"
+        -> "hired"
+        -> "rejected"
     """
     applied = "applied"
     cv_screened = "cv_screened"
@@ -29,7 +39,11 @@ class CandidateStatus(str, enum.Enum):
     voice_passed = "voice_passed"
     voice_rejected = "voice_rejected"
     interview_scheduled = "interview_scheduled"
+    interview_passed = "interview_passed"
+    interview_rejected = "interview_rejected"
     decision_made = "decision_made"
+    hired = "hired"
+    rejected = "rejected"
 
 
 class InterviewStatus(str, enum.Enum):

start.sh

@@ -4,6 +4,74 @@ set -e
 # Hugging Face provides PORT; default to 7860 locally
 export PORT="${PORT:-7860}"
 
+# Locate PostgreSQL binaries (initdb/pg_ctl) even on versioned paths
+PG_BIN=$(dirname $(find /usr/lib/postgresql -name initdb | head -n 1 2>/dev/null))
+if [ -n "$PG_BIN" ]; then
+  export PATH="$PG_BIN:$PATH"
+fi
+
+# Normalize DB host: if set to 'db' (compose style), force localhost for single-container run
+if [ "${POSTGRES_HOST}" = "db" ] || [ "${POSTGRES_HOST}" = "\"db\"" ] || [ -z "${POSTGRES_HOST}" ]; then
+  export POSTGRES_HOST="127.0.0.1"
+fi
+export POSTGRES_PORT="${POSTGRES_PORT:-5432}"
+export POSTGRES_USER="${POSTGRES_USER:-agentic_user}"
+export POSTGRES_PASSWORD="${POSTGRES_PASSWORD:-password123}"
+export POSTGRES_DB="${POSTGRES_DB:-agentic_hr}"
+
+echo "[start.sh] PORT=${PORT}"
+echo "[start.sh] POSTGRES_HOST=${POSTGRES_HOST}"
+echo "[start.sh] POSTGRES_PORT=${POSTGRES_PORT}"
+echo "[start.sh] POSTGRES_USER=${POSTGRES_USER}"
+echo "[start.sh] POSTGRES_DB=${POSTGRES_DB}"
+
+# Start local Postgres if not already running
+export PGDATA=/var/lib/postgresql/data
+mkdir -p "$PGDATA"
+chown -R postgres:postgres "$PGDATA"
+mkdir -p /var/run/postgresql
+chown postgres:postgres /var/run/postgresql
+
+if [ ! -s "$PGDATA/PG_VERSION" ]; then
+  echo "[start.sh] Initializing postgres data dir..."
+  gosu postgres initdb -D "$PGDATA"
+fi
+
+echo "[start.sh] Starting postgres on port ${POSTGRES_PORT}..."
+if ! gosu postgres pg_ctl -D "$PGDATA" -o "-p ${POSTGRES_PORT} -k /var/run/postgresql" -w start >> /var/log/postgres.log 2>&1; then
+  echo "[start.sh] Postgres failed to start. Last log lines:"
+  tail -n 100 /var/log/postgres.log || true
+  exit 1
+fi
+echo "[start.sh] Postgres started."
+echo "[start.sh] Postgres last log lines:"
+tail -n 50 /var/log/postgres.log || true
+
+# Ensure user/db exist
+gosu postgres psql -h 127.0.0.1 -p "${POSTGRES_PORT}" -v ON_ERROR_STOP=1 --command "DO \$\$
+BEGIN
+   IF NOT EXISTS (SELECT FROM pg_catalog.pg_roles WHERE rolname = '${POSTGRES_USER}') THEN
+      CREATE ROLE ${POSTGRES_USER} LOGIN PASSWORD '${POSTGRES_PASSWORD}';
+   END IF;
+END
+\$\$;" || true
+gosu postgres psql -h 127.0.0.1 -p "${POSTGRES_PORT}" -v ON_ERROR_STOP=1 --command "CREATE DATABASE ${POSTGRES_DB} OWNER ${POSTGRES_USER}" || true
+echo "[start.sh] Postgres user/db ensured."
+
+# Create tables if they don't exist
+echo "[start.sh] Ensuring database tables exist..."
+python - <<'PY'
+import os
+from src.database.candidates.models import Base
+from src.database.candidates.client import engine
+
+try:
+    Base.metadata.create_all(bind=engine)
+    print("[db-init] Tables ensured.")
+except Exception as e:
+    print(f"[db-init] Failed to create tables: {e}")
+PY
+
 # Defaults for local in-container routing; can be overridden via env
 export SUPERVISOR_API_URL="${SUPERVISOR_API_URL:-http://127.0.0.1:8080/api/v1/supervisor}"
 export DATABASE_API_URL="${DATABASE_API_URL:-http://127.0.0.1:8080/api/v1/db}"
@@ -16,4 +84,4 @@ uvicorn src.api.app:app --host 0.0.0.0 --port 8080 &
 sleep 2
 
 # Run Gradio frontend
-python src/frontend/gradio/app.py
+python src/frontend/gradio/app.py

tests/create_dummy_candidate.py

@@ -1,8 +1,8 @@
 import uuid
 from datetime import datetime
-from src.database.candidates.client import SessionLocal
-from src.database.candidates.models import Candidate, CVScreeningResult
-from src.state.candidate import CandidateStatus
+from src.backend.database.candidates.client import SessionLocal
+from src.backend.database.candidates.models import Candidate, CVScreeningResult
+from src.backend.state.candidate import CandidateStatus
 
 def create_dummy_candidate():
     with SessionLocal() as db:

tests/verify_voice_integration.py

@@ -10,9 +10,9 @@ load_dotenv()
 # Add src to path
 sys.path.append(os.path.join(os.path.dirname(__file__), ".."))
 
-from src.database.candidates.models import Candidate, CVScreeningResult, Base
-from src.database.candidates.client import SessionLocal, engine
-from src.agents.voice_screening.utils.questions import get_screening_questions
+from src.backend.database.candidates.models import Candidate, CVScreeningResult, Base
+from src.backend.database.candidates.client import SessionLocal, engine
+from src.backend.agents.voice_screening.utils.questions import get_screening_questions
 
 def verify_integration():
     print("Verifying integration...")