feat: auto-create PromptConfig for customPrompt + filtered extraction example

## Bug Fix
- Auto-create PromptConfig when customPrompt is used without promptConfigId
- Fixes FK constraint violation error during ingestion
## New Example
- Add filtered-extraction-demo.ts demonstrating selective extraction
- Extract only TEXT, QUESTION, LIST, TABLE with customPrompt
- Context enrichment only for TEXT chunks (cost optimization)
## Documentation
- Add "Custom Prompt / Filtered Extraction" section to README.md
- Update examples/README.md with new demo instructions
## Code Quality (previous commits)
- Add GeminiAPIError, PDFProcessingError, ContentPolicyError
- Replace console.log/warn with structured logger
- Fix any type casts in gemini.service.ts and ingestion.engine.ts
All tests pass (59/59), TypeScript check clean.

Author: msbayindir

README.md

@@ -276,6 +276,56 @@ const rag = new ContextRAG({
 
 ---
 
+## 🎯 Custom Prompt / Filtered Extraction
+
+Extract only specific content types without going through the Discovery flow:
+
+```typescript
+// Extract ONLY specific types with custom prompt
+const result = await rag.ingest({
+  file: './book.pdf',
+  customPrompt: `
+    Extract ONLY these content types:
+    - TEXT: Normal paragraphs
+    - QUESTION: Multiple choice questions
+    - LIST: Bulleted or numbered lists
+    - TABLE: Data tables
+
+    SKIP these types:
+    - HEADING, CODE, QUOTE, IMAGE_REF
+  `,
+  // Context enrichment only for TEXT chunks (cost optimization)
+  // Configure via ragEnhancement.skipChunkTypes
+});
+```
+
+### Configuration for Selective Context Enrichment
+
+```typescript
+const rag = new ContextRAG({
+  prisma,
+  geminiApiKey: process.env.GEMINI_API_KEY!,
+  
+  ragEnhancement: {
+    approach: 'anthropic_contextual',
+    strategy: 'llm',
+    // Only TEXT chunks get context enrichment
+    // Other types (TABLE, LIST, QUESTION) are extracted but not enriched
+    skipChunkTypes: ['HEADING', 'IMAGE_REF', 'TABLE', 'CODE', 'QUOTE', 'MIXED', 'QUESTION', 'LIST'],
+  },
+});
+
+// PromptConfig is auto-created when using customPrompt
+await rag.ingest({
+  file: './document.pdf',
+  customPrompt: 'Your custom extraction instructions...',
+});
+```
+
+> **Note:** When using `customPrompt` without `promptConfigId`, the system automatically creates a PromptConfig for you.
+
+---
+
 ## ⚙️ Configuration
 
 ```typescript

examples/README.md

@@ -42,11 +42,28 @@
    cp /path/to/your/document.pdf examples/test.pdf
    ```
 
-2. Run the demo:
+2. Run the full demo:
    ```bash
    npx tsx examples/demo.ts
    ```
 
+3. **NEW:** Run the filtered extraction demo:
+   ```bash
+   npx tsx examples/filtered-extraction-demo.ts
+   ```
+
+### Filtered Extraction Demo
+
+The `filtered-extraction-demo.ts` demonstrates:
+- **Custom Prompt:** Extract only TEXT, QUESTION, LIST, TABLE (skip HEADING, CODE, etc.)
+- **Selective Context:** Context enrichment only for TEXT chunks (cost optimization)
+- **Search Filtering:** Query only specific chunk types
+
+This is useful when you want to:
+- Extract specific content types from a document
+- Reduce context generation costs by skipping non-essential chunks
+- Focus RAG search on particular content categories
+
 ## Expected Output
 
 ```

examples/filtered-extraction-demo.ts

@@ -52,7 +52,7 @@ async function main() {
     const rag = new ContextRAG({
         prisma,
         geminiApiKey: process.env.GEMINI_API_KEY!,
-        model: 'gemini-2.5-flash',
+        model: 'gemini-3-pro-preview',
         generationConfig: {
             temperature: 0.2,
             maxOutputTokens: 16384,
@@ -71,8 +71,8 @@ async function main() {
         },
 
         batchConfig: {
-            pagesPerBatch: 10,
-            maxConcurrency: 2,
+            pagesPerBatch: 30,
+            maxConcurrency: 3,
         },
         logging: {
             level: 'info',

src/engines/ingestion.engine.ts

@@ -135,11 +135,75 @@ export class IngestionEngine {
                 .split('\n')
                 .map(line => line.trim())
                 .filter(line => line.length > 0);
+
+            // Auto-create PromptConfig if customPrompt is provided but no promptConfigId
+            if (!promptConfigId) {
+                const customDocType = options.documentType ?? 'CustomPrompt';
+                const existingConfig = await this.promptConfigRepo.getDefault(customDocType);
+
+                if (existingConfig) {
+                    // Use existing config
+                    promptConfigId = existingConfig.id;
+                    this.logger.debug('Using existing PromptConfig for customPrompt', {
+                        promptConfigId,
+                        documentType: customDocType
+                    });
+                } else {
+                    // Create new PromptConfig for custom prompt
+                    const newConfig = await this.promptConfigRepo.create({
+                        documentType: customDocType,
+                        name: `Custom Extraction - ${new Date().toISOString().slice(0, 10)}`,
+                        systemPrompt: options.customPrompt,
+                        chunkStrategy: {
+                            maxTokens: 800,
+                            splitBy: 'semantic',
+                            preserveTables: true,
+                            preserveLists: true,
+                        },
+                        setAsDefault: true,
+                        changeLog: 'Auto-created from customPrompt',
+                    });
+                    promptConfigId = newConfig.id;
+                    this.logger.info('Created new PromptConfig for customPrompt', {
+                        promptConfigId,
+                        documentType: customDocType
+                    });
+                }
+            }
         } else if (documentInstructions.length === 0) {
             documentInstructions = DEFAULT_DOCUMENT_INSTRUCTIONS
                 .split('\n')
                 .map(line => line.replace(/^-\s*/, '').trim())
                 .filter(line => line.length > 0);
+
+            // Create default PromptConfig if none exists
+            if (!promptConfigId) {
+                const defaultDocType = options.documentType ?? 'General';
+                const existingConfig = await this.promptConfigRepo.getDefault(defaultDocType);
+
+                if (existingConfig) {
+                    promptConfigId = existingConfig.id;
+                } else {
+                    const newConfig = await this.promptConfigRepo.create({
+                        documentType: defaultDocType,
+                        name: 'Default Extraction',
+                        systemPrompt: DEFAULT_DOCUMENT_INSTRUCTIONS,
+                        chunkStrategy: {
+                            maxTokens: 800,
+                            splitBy: 'semantic',
+                            preserveTables: true,
+                            preserveLists: true,
+                        },
+                        setAsDefault: true,
+                        changeLog: 'Auto-created as system default',
+                    });
+                    promptConfigId = newConfig.id;
+                    this.logger.info('Created default PromptConfig', {
+                        promptConfigId,
+                        documentType: defaultDocType
+                    });
+                }
+            }
         }
 
         // Create batches