Remove local keyword heuristics and replace with clean error strings

README.md

@@ -25,8 +25,7 @@ Hugging Face Space: [build-small-hackathon/innerspace](https://huggingface.co/sp
 * **Calming Mindful Interface**: A customized, premium dark-slate/violet visual dashboard optimized for reflection.
 * **CBT Reflector Coach**: Engages the writer with open-ended, therapeutic prompts to explore underlying thoughts.
 * **Cognitive Distortion Scanner**: Highlights thinking patterns like *Catastrophizing*, *All-or-Nothing Thinking*, or *Should Statements* to raise cognitive awareness.
-* **Hybrid Inference**: Run locally on CUDA GPU (ZeroGPU space) with serverless API fallbacks and local rule-based keyword heuristics.
-* **Zero External Dependencies (Offline Heuristics)**: Fully functional without an internet connection or Hugging Face tokens using local keyword parsing.
+* **Hybrid Inference**: Run locally on CUDA GPU (ZeroGPU space) with serverless API fallbacks.
 
 ---
 
@@ -60,19 +59,13 @@ The application is built around single-responsibility layers following SOLID pri
 | - ZeroGPU A10G  |                 | - Section     |
 | - HF API Client |                 |   Splitting   |
 +-----------------+                 +---------------+
-      | (If Inference Fails)
-      v
-+-----+-----+-----+
-| Heuristic Engine|
-| (heuristics.py) | <--- Local offline keyword analyzer for sentiment/CBT fallbacks
-+-----------------+
 ```
 
 1. **User Input**: The user writes a journal entry in the writing block or uploads a `.txt`/`.md` entry.
 2. **Core Routing**: The click event invokes `analyze_journal_ui` through [core.py](core.py).
 3. **Model Generation**: The prompt is processed by the 1.2B parameter OpenBMB model in `inference.py`. It runs on a GPU locally when available.
 4. **API Fallback**: If the local GPU/CPU is busy or unavailable, the system transparently falls back to the Hugging Face Serverless API.
-5. **Heuristic Fallback**: If both inference paths fail (e.g., completely offline with no token), the system falls back to rule-based keyword matchers in `heuristics.py` to extract basic emotions and distortions, ensuring the UI remains active and useful.
+5. **Error Handling**: If both inference paths fail (e.g., completely offline with no token), the system returns structured analysis-unavailable error indicators in the dashboard.
 
 ---
 
@@ -127,7 +120,6 @@ The project directory contains:
 - [analyzer.py](analyzer.py) - Analysis orchestrator and ZeroGPU wrapper.
 - [inference.py](inference.py) - Lazy model loader and local/remote text generators.
 - [parser.py](parser.py) - IO reader and section separator.
-- [heuristics.py](heuristics.py) - Keyword match fallback system.
 - [ui.py](ui.py) - Gradio block layout and interaction hooks.
 - [styles.py](styles.py) - Calming violet custom styling overrides.
 - [tune_journal.py](tune_journal.py) - Modal fine-tuning script.

analyzer.py

@@ -29,12 +29,6 @@ except ImportError:
 from config import ENTRY_LIMIT, MODEL_ID, PARAMETER_COUNT
 from inference import run_model_inference
 from parser import extract_journal_text, parse_sections
-from heuristics import (
-    build_sentiment_fallback,
-    build_areas_fallback,
-    build_distortions_fallback,
-    build_reflection_fallback,
-)
 
 
 @dataclass(frozen=True)
@@ -105,23 +99,14 @@ def analyze_journal(file_path: str | None, raw_text: str) -> JournalReport:
         ]
     )
 
-    # Route output parsing: split sections if model succeeded, otherwise execute heuristics
+    # Route output parsing: split sections if model succeeded, otherwise return error strings
     if response.strip():
         sentiment, areas, distortions, reflection = parse_sections(response)
-        # Apply heuristics fallback if model failed to populate specific sections
-        if "not resolved" in sentiment:
-            sentiment = build_sentiment_fallback(trimmed_entry)
-        if "not resolved" in areas:
-            areas = build_areas_fallback(trimmed_entry)
-        if "not resolved" in distortions:
-            distortions = build_distortions_fallback(trimmed_entry)
-        if reflection.strip() == "How are you feeling about these thoughts today?":
-            reflection = build_reflection_fallback(trimmed_entry)
     else:
-        sentiment = build_sentiment_fallback(trimmed_entry)
-        areas = build_areas_fallback(trimmed_entry)
-        distortions = build_distortions_fallback(trimmed_entry)
-        reflection = build_reflection_fallback(trimmed_entry)
+        sentiment = "- Analysis unavailable"
+        areas = "- Analysis unavailable"
+        distortions = "- Analysis unavailable"
+        reflection = "An error occurred during model analysis. Please check your network connection or Hugging Face access token."
 
     return JournalReport(
         entry_text=trimmed_entry,

heuristics.py

@@ -1,100 +0,0 @@
-"""
-Module containing keyword-based fallback heuristics.
-Analyzes text patterns to extract mood, areas of concern, distortions, and questions.
-"""
-
-from __future__ import annotations
-
-
-def build_sentiment_fallback(text: str) -> str:
-    """Detects simple emotion keywords to build a fallback mood classification."""
-    lowered = text.lower()
-    emotions = []
-
-    if any(w in lowered for w in ["sad", "down", "cry", "hurt", "grief", "lonely"]):
-        emotions.append("- Sadness / Loneliness")
-    if any(
-        w in lowered for w in ["angry", "mad", "annoy", "frustrate", "hate", "irritate"]
-    ):
-        emotions.append("- Anger / Frustration")
-    if any(
-        w in lowered
-        for w in ["happy", "glad", "great", "excite", "joy", "peace", "love"]
-    ):
-        emotions.append("- Joy / Contentment")
-    if any(
-        w in lowered for w in ["anxious", "worry", "afraid", "scare", "fear", "nervous"]
-    ):
-        emotions.append("- Anxiety / Concern")
-
-    if not emotions:
-        return "- Neutral / Reflective mood"
-    return "\n".join(emotions)
-
-
-def build_areas_fallback(text: str) -> str:
-    """Maps daily concerns to specific life areas based on vocabulary patterns."""
-    lowered = text.lower()
-    areas = []
-
-    if any(w in lowered for w in ["work", "job", "office", "career", "boss", "task"]):
-        areas.append("- Career & Work")
-    if any(
-        w in lowered
-        for w in [
-            "family",
-            "mom",
-            "dad",
-            "kid",
-            "child",
-            "wife",
-            "husband",
-            "friend",
-            "relationship",
-        ]
-    ):
-        areas.append("- Relationships & Social")
-    if any(
-        w in lowered for w in ["health", "sick", "doctor", "gym", "run", "eat", "body"]
-    ):
-        areas.append("- Health & Wellness")
-    if any(w in lowered for w in ["money", "pay", "buy", "budget", "cost", "finances"]):
-        areas.append("- Finances")
-
-    if not areas:
-        return "- Personal Growth & General Life"
-    return "\n".join(areas)
-
-
-def build_distortions_fallback(text: str) -> str:
-    """Detects common cognitive distortions by analyzing linguistic patterns (should, catastrophizing, etc.)."""
-    lowered = text.lower()
-    distortions = []
-
-    if any(w in lowered for w in ["always", "never", "nothing", "everything"]):
-        distortions.append(
-            "- All-or-Nothing Thinking (viewing things in black-and-white)"
-        )
-    if any(
-        w in lowered
-        for w in ["ruined", "disaster", "fail", "worst", "end of the world"]
-    ):
-        distortions.append("- Catastrophizing (expecting the worst outcome)")
-    if any(w in lowered for w in ["must", "should", "ought"]):
-        distortions.append("- 'Should' Statements (unrealistic self-imposed rules)")
-
-    if not distortions:
-        return "- None detected (heuristics analysis)"
-    return "\n".join(distortions)
-
-
-def build_reflection_fallback(text: str) -> str:
-    """Generates a simple, comforting CBT reflection question matching dominant keywords."""
-    lowered = text.lower()
-
-    if "work" in lowered or "job" in lowered:
-        return "What would a small, manageable step towards easing this work pressure look like today?"
-    if "always" in lowered or "never" in lowered:
-        return "You mentioned things 'always' or 'never' going this way. Can you think of a single exception?"
-
-    return "What is one kind thing you can say to yourself about these thoughts today?"