refactor: layered architecture (use_cases/queries/domain) + UI theme Release 2026.8.9: updated UI (light theme default, dark-mode toggle, ember accent), theory answer evaluation scoring, Whisper improvements. Removes the services layer in favor of use_cases/queries/domain and shared infrastructure gateways; updates CHANGELOG, README, ARCHITECTURE.

.gitignore

@@ -46,6 +46,8 @@ env/
 
 # Database
 *.db
+*.db-shm
+*.db-wal
 *.sqlite
 *.sqlite3
 data/config.json

CHANGELOG.md

@@ -10,10 +10,24 @@ Work in progress is accumulated under `[Unreleased]`; on release, that section b
 
 ### Changed
 
+### Fixed
+
+### Removed
+
+## 2026.8.9
+
+### Added
+
+### Changed
+
+- **Updated UI** — light theme is now the default, with a **dark theme toggle** (sun/moon) in the navigation bar. Your choice is remembered and falls back to your system preference on first visit. The whole palette moved to a warm, restrained **«ember»** (orange/red) accent with better contrast throughout
 - **Theory answer evaluation** — load `expected_points` rubric bullets from question banks, pass them through evaluation prompts with explicit candidate-only scoring rules, and use temperature 0 for structured LLM evaluation
 
 ### Fixed
 
+- **Coding timer** — when a coding round timer expires, the round now submits automatically and the session advances even if you refresh the page
+- **Whisper transcription** — more robust audio transcription (voice-activity detection disabled) with clearer audio-answer logging
+
 ### Removed
 
 ## 2026.7.14

README.md

@@ -2,7 +2,7 @@
 
 [![Python 3.12+](https://img.shields.io/badge/python-3.12+-blue.svg)](https://www.python.org/downloads/)
 [![License: Apache 2.0](https://img.shields.io/badge/License-Apache%202.0-yellow.svg)](https://opensource.org/licenses/Apache-2.0)
-[![Version](https://img.shields.io/badge/version-2026.6.12-blue.svg)](CHANGELOG.md)
+[![Version](https://img.shields.io/badge/version-2026.8.9-blue.svg)](CHANGELOG.md)
 
 Open-source AI technical interview trainer. Practice **theory Q&A**, **live coding**, or **both in one session** from curated YAML banks — with structured scoring, follow-ups, optional voice, and a local results history. Bring your own LLM (cloud or local).
 
@@ -34,30 +34,6 @@ A general chat assistant is flexible, but it does not run an **interview** for y
 https://github.com/user-attachments/assets/25655f1e-89d3-472f-8c1f-3f154df622b2
 
 
-**Dashboard** — recent sessions and quick start
-
-<p align="center">
-  <img src="./assets/dashboard.png" alt="GrillKit dashboard" width="900" />
-</p>
-
-**Interview setup** — question-bank tracks, levels, topics, and session options
-
-<p align="center">
-  <img src="./assets/interview-setup.png" alt="Interview setup" width="900" />
-</p>
-
-**Coding section** — Monaco editor, Run on public tests, Submit for AI evaluation
-
-<p align="center">
-  <img src="./assets/coding.png" alt="Coding interview session" width="900" />
-</p>
-
-**Theory section** — real-time Q&A with AI scoring and final evaluation
-
-<p align="center">
-  <img src="./assets/interview-session.png" alt="Completed interview with evaluation" width="900" />
-</p>
-
 ## Features
 
 ### Session modes
@@ -85,6 +61,7 @@ Coding modes need a running [Judge0](https://github.com/judge0/judge0) instance
 - **Known questions** — mark theory or coding bank items as **I know this** during an interview or on review pages; optionally exclude them on **New interview** setup; manage the list at `/known-questions/manage`
 - **Dashboard** — recent sessions on the home page (completed sessions link to results)
 - **Setup** — model catalog on `/config`, interview locale, Whisper/Piper downloads from the UI
+- **Theme** — light theme by default with a **dark mode toggle** (sun/moon) in the navbar; your choice is remembered and follows your system preference on first visit
 - **Deployment** — Docker Compose on port 8000 with `./data` volume for config, DB, and models
 
 ## Quick start

app/ai/__init__.py

@@ -6,14 +6,22 @@
 including base classes, factory methods, and concrete implementations.
 """
 
-from app.ai.base import AIProvider, GenerationResult, Message
+from app.ai.base import (
+    AIProvider,
+    AudioCapableProvider,
+    GenerationResult,
+    Message,
+    StreamingProvider,
+)
 from app.ai.factory import ProviderFactory
 from app.ai.openai_compatible import OpenAICompatibleProvider
 
 __all__ = [
     "AIProvider",
+    "AudioCapableProvider",
     "GenerationResult",
     "Message",
     "OpenAICompatibleProvider",
     "ProviderFactory",
+    "StreamingProvider",
 ]

app/ai/audio_probe.py

@@ -3,6 +3,8 @@
 """Minimal WAV payloads for audio capability probes."""
 
 import io
+import math
+import struct
 import wave
 
 from app.shared.infrastructure.audio_wav import CANONICAL_AUDIO_SAMPLE_RATE_HZ
@@ -13,23 +15,32 @@
 def minimal_wav_bytes(
     *,
     sample_rate: int = CANONICAL_AUDIO_SAMPLE_RATE_HZ,
-    duration_sec: float = 0.1,
+    duration_sec: float = 0.5,
+    tone_freq_hz: float = 440.0,
 ) -> bytes:
-    """Build a short silent mono PCM WAV for connection testing.
+    """Build a short mono PCM WAV beep for connection testing.
+
+    A pure-silence clip is rejected by most multimodal audio models (they
+    require audible speech), so the probe sends a short audible tone instead.
 
     Args:
         sample_rate: Sample rate in Hz.
-        duration_sec: Duration of silence in seconds.
+        duration_sec: Duration of the tone in seconds.
+        tone_freq_hz: Frequency of the probe tone in Hz.
 
     Returns:
         WAV file bytes suitable for provider audio probes.
     """
     frame_count = max(1, int(sample_rate * duration_sec))
-    pcm = b"\x00\x00" * frame_count
+    amplitude = 0.2  # moderate volume, well below clipping
+    pcm = bytearray()
+    for i in range(frame_count):
+        sample = amplitude * math.sin(2 * math.pi * tone_freq_hz * i / sample_rate)
+        pcm += struct.pack("<h", int(sample * 32767))
     buffer = io.BytesIO()
     with wave.open(buffer, "wb") as wav_file:
         wav_file.setnchannels(1)
         wav_file.setsampwidth(2)
         wav_file.setframerate(sample_rate)
-        wav_file.writeframes(pcm)
+        wav_file.writeframes(bytes(pcm))
     return buffer.getvalue()

app/ai/base.py

@@ -1,9 +1,10 @@
 # Copyright 2026 GrillKit Contributors
 # SPDX-License-Identifier: Apache-2.0
-"""Abstract base class for AI providers.
+"""Abstract base class and capability protocols for AI providers.
 
 This module defines the base abstractions and data models for AI providers,
-including message structures, generation results, and the abstract provider interface.
+including message structures, generation results, and capability-based
+provider interfaces (ISP).
 """
 
 from abc import ABC, abstractmethod
@@ -40,14 +41,10 @@ class GenerationResult:
 
 
 class AIProvider(ABC):
-    """Abstract base for all AI providers.
+    """Base text-generation provider interface.
 
-    This class defines the interface that all AI providers must implement,
-    including generation, streaming, validation, and metadata.
-
-    Attributes:
-        model: The model name/identifier to use.
-        config: Additional provider-specific configuration.
+    All providers must implement at least text-based generation and lifecycle
+    methods.  Streaming and audio are opt-in via separate capability protocols.
     """
 
     def __init__(self, model: str, **kwargs: object) -> None:
@@ -64,29 +61,10 @@ def __init__(self, model: str, **kwargs: object) -> None:
     @abstractmethod
     def name(self) -> str:
         """Provider display name."""
-        pass
-
-    @abstractmethod
-    def supports_streaming(self) -> bool:
-        """Check if provider supports streaming."""
-        pass
 
     @abstractmethod
     async def validate(self) -> bool:
         """Validate API key and connection."""
-        pass
-
-    @abstractmethod
-    async def probe_audio_input(self, audio_wav: bytes) -> bool:
-        """Probe whether the endpoint accepts multimodal audio input.
-
-        Args:
-            audio_wav: Canonical WAV bytes (mono PCM).
-
-        Returns:
-            True when the provider accepts the audio probe request.
-        """
-        pass
 
     @abstractmethod
     async def generate(
@@ -95,71 +73,56 @@ async def generate(
         temperature: float = 0.7,
         max_tokens: int = 2000,
     ) -> GenerationResult:
-        """Generate a single response.
+        """Generate a single response."""
 
-        Args:
-            messages: List of conversation messages.
-            temperature: Sampling temperature (0.0 to 2.0).
-            max_tokens: Maximum tokens to generate.
+    @abstractmethod
+    async def close(self) -> None:
+        """Close the provider and release resources."""
 
-        Returns:
-            The generation result with content and metadata.
 
-        Raises:
-            ValueError: If the request fails or parameters are invalid.
-        """
-        pass
+class StreamingProvider(ABC):
+    """Capability protocol for providers that support token streaming."""
 
     @abstractmethod
-    async def generate_with_audio(
+    def supports_streaming(self) -> bool:
+        """Check if provider supports streaming."""
+
+    @abstractmethod
+    def generate_stream(
         self,
         messages: list[Message],
-        audio_wav: bytes,
-        *,
-        user_text: str,
         temperature: float = 0.7,
         max_tokens: int = 2000,
-    ) -> GenerationResult:
-        """Generate a response from system messages, user text, and audio.
-
-        Args:
-            messages: System (and optional assistant) messages without user audio.
-            audio_wav: Canonical WAV bytes representing the user's spoken answer.
-            user_text: Text context for the user turn (question prompt, no answer text).
-            temperature: Sampling temperature (0.0 to 2.0).
-            max_tokens: Maximum tokens to generate.
-
-        Returns:
-            The generation result with content and metadata.
+    ) -> AsyncIterator[str]:
+        """Stream response tokens.
 
-        Raises:
-            ValueError: If the request fails or parameters are invalid.
+        Yields:
+            Chunks of generated text as they become available.
         """
-        pass
 
-    @abstractmethod
-    async def close(self) -> None:
-        """Close the provider and release resources."""
-        pass
+
+class AudioCapableProvider(ABC):
+    """Capability protocol for providers that accept multimodal audio input."""
 
     @abstractmethod
-    def generate_stream(
+    async def generate_with_audio(
         self,
         messages: list[Message],
+        audio_wav: bytes,
+        *,
+        user_text: str,
         temperature: float = 0.7,
         max_tokens: int = 2000,
-    ) -> AsyncIterator[str]:
-        """Stream response tokens.
+    ) -> GenerationResult:
+        """Generate a response from system messages, user text, and audio."""
 
-        Args:
-            messages: List of conversation messages.
-            temperature: Sampling temperature (0.0 to 2.0).
-            max_tokens: Maximum tokens to generate.
+    @abstractmethod
+    async def probe_audio_input(self, audio_wav: bytes) -> bool:
+        """Probe whether the endpoint accepts multimodal audio input.
 
-        Yields:
-            Chunks of generated text as they become available.
+        Args:
+            audio_wav: Canonical WAV bytes (mono PCM).
 
-        Raises:
-            ValueError: If the request fails or parameters are invalid.
+        Returns:
+            True when the provider accepts the audio probe request.
         """
-        pass

app/ai/faster_whisper_transcriber.py

@@ -52,7 +52,7 @@ def _transcribe() -> str:
             result = "".join((segment.text or "") for segment in segment_list).strip()
             logger.info(
                 "Whisper transcript: language=%s segments=%d result=%r",
-                info.language,
+                getattr(info, "language", language) if info is not None else language,
                 len(segment_list),
                 result,
             )