The legal co-pilot backend is live and operational.
\n \n \n `);\n});\n\n// Health check endpoint\napp.get('/api/health', (req, res) => {\n res.json({\n status: 'healthy',\n timestamp: new Date().toISOString(),\n uptime: process.uptime(),\n memory: process.memoryUsage(),\n rag: {\n loaded: documentChunks.length > 0,\n chunks: documentChunks.length\n },\n ollama: {\n baseUrl: OLLAMA_BASE_URL,\n embeddingModel: EMBEDDING_MODEL,\n chatModel: CHAT_MODEL\n }\n });\n});\n\n// Get RAG statistics\napp.get('/api/stats', (req, res) => {\n const sources = [...new Set(documentChunks.map(c => c.source))];\n res.json({\n totalChunks: documentChunks.length,\n totalSources: sources.length,\n sources: sources,\n memoryUsage: process.memoryUsage()\n });\n});\n\n// Voice Configuration Endpoint (tells client which STT methods are available)\napp.get('/api/voice/config', (req, res) => {\n // Check if Bhashini credentials are properly configured (not empty or placeholder values)\n const isPlaceholder = (value) => {\n if (!value || typeof value !== 'string') return true;\n const trimmed = value.trim().toLowerCase();\n return (\n trimmed === '' ||\n trimmed === 'your_bhashini_api_key_here' ||\n trimmed === 'your_bhashini_user_id_here' ||\n trimmed.startsWith('your_') ||\n trimmed === 'placeholder'\n );\n };\n\n const bhashiniConfigured = !isPlaceholder(BHASHINI_API_KEY) && !isPlaceholder(BHASHINI_USER_ID);\n \n // Only return pipelineId when Bhashini is actually configured (avoid info leak)\n const bhashiniResponse = { available: bhashiniConfigured };\n if (bhashiniConfigured) {\n bhashiniResponse.pipelineId = BHASHINI_PIPELINE_ID;\n }\n \n res.json({\n bhashini: bhashiniResponse\n // Web Speech API availability is checked client-side\n });\n});\n\n// Bhashini Voice Processing Endpoint (STT & TTS Proxy)\napp.post('/api/voice/process', async (req, res) => {\n try {\n const { task, audioContent, text, sourceLanguage, targetLanguage } = req.body;\n \n // Check if Bhashini credentials are provided, otherwise enter Mock Mode\n if (!BHASHINI_API_KEY || !BHASHINI_USER_ID) {\n console.log('⚠️ Bhashini Credentials missing. Running in MOCK MODE.');\n \n if (task === 'asr') {\n // Mock transcription\n return res.json({ \n status: 'mock',\n output: [{ source: 'JusticeAI Mock ASR', transcription: 'This is a simulated transcription because Bhashini API keys are not configured.' }] \n });\n } else if (task === 'tts') {\n // Mock TTS (no audio)\n return res.json({ \n status: 'mock',\n output: [{ audioContent: '', message: 'Simulated TTS output' }] \n });\n }\n }\n\n // Step 1: Fetch Pipeline Configuration\n const configPayload = {\n pipelineTasks: [\n {\n taskType: task, // 'asr' or 'tts'\n config: {\n language: {\n sourceLanguage: sourceLanguage || 'hi'\n }\n }\n }\n ],\n pipelineConfig: {\n pipelineId: BHASHINI_PIPELINE_ID\n }\n };\n\n const configResponse = await fetch(`${BHASHINI_BASE_URL}/config`, {\n method: 'POST',\n headers: {\n 'Content-Type': 'application/json',\n 'ulcaApiKey': BHASHINI_API_KEY,\n 'userID': BHASHINI_USER_ID\n },\n body: JSON.stringify(configPayload)\n });\n\n if (!configResponse.ok) {\n const errorText = await configResponse.text();\n throw new Error(`Bhashini Config Error: ${configResponse.status} - ${errorText}`);\n }\n\n const configData = await configResponse.json();\n\n // Step 2: Compute Inference\n const computePayload = {\n pipelineTasks: [\n {\n taskType: task,\n config: configData.pipelineResponseConfig[0].config,\n input: task === 'asr' \n ? [{ audioContent }] \n : [{ source: text }]\n }\n ],\n pipelineResponseConfig: configData.pipelineResponseConfig\n };\n\n const computeResponse = await fetch(`${BHASHINI_BASE_URL}/compute`, {\n method: 'POST',\n headers: {\n 'Content-Type': 'application/json',\n 'Authorization': configData.pipelineInferenceAPIEndPoint.inferenceApiKey.value,\n 'Accept': '*/*'\n },\n body: JSON.stringify(computePayload)\n });\n\n if (!computeResponse.ok) {\n const errorText = await computeResponse.text();\n throw new Error(`Bhashini Compute Error: ${computeResponse.status} - ${errorText}`);\n }\n\n const computeData = await computeResponse.json();\n \n // Return the specific output based on task\n if (task === 'asr') {\n res.json({\n transcription: computeData.pipelineResponse[0].output[0].source,\n raw: computeData\n });\n } else {\n res.json({\n audioContent: computeData.pipelineResponse[0].audio[0].audioContent,\n raw: computeData\n });\n }\n\n } catch (err) {\n console.error('Bhashini Proxy Error:', err);\n res.status(500).json({ error: err.message });\n }\n});\n\n// Upload documents endpoint\napp.post('/api/upload', upload.array('documents', 5), async (req, res) => {\n try {\n if (!req.files || req.files.length === 0) {\n return res.status(400).json({ error: 'No files uploaded' });\n }\n\n const uploadedFiles = [];\n const failedFiles = [];\n\n for (const file of req.files) {\n try {\n const filePath = file.path;\n const fileName = file.originalname;\n const ext = path.extname(fileName).toLowerCase();\n let text = '';\n\n if (ext === '.pdf') {\n const dataBuffer = fs.readFileSync(filePath);\n const pdfData = await parsePdf(dataBuffer);\n text = pdfData.text;\n } else if (ext === '.txt' || ext === '.md') {\n text = fs.readFileSync(filePath, 'utf-8');\n } else {\n failedFiles.push({ name: fileName, error: 'Unsupported file type' });\n continue;\n }\n\n // Chunk and embed\n const rawChunks = splitTextIntoChunks(text, 1000, 200);\n let chunksEmbedded = 0;\n\n for (const chunk of rawChunks) {\n const vector = await embeddings.embedQuery(chunk);\n documentChunks.push({\n content: chunk,\n vector: vector,\n source: fileName\n });\n chunksEmbedded++;\n }\n\n uploadedFiles.push({\n name: fileName,\n size: file.size,\n chunks: chunksEmbedded\n });\n\n // Clean up uploaded file\n fs.unlinkSync(filePath);\n\n } catch (err) {\n failedFiles.push({ name: file.originalname, error: err.message });\n }\n }\n\n res.json({\n message: 'Documents processed',\n uploaded: uploadedFiles,\n failed: failedFiles,\n totalChunks: documentChunks.length\n });\n\n } catch (error) {\n console.error('Upload error:', error);\n res.status(500).json({ error: error.message });\n }\n});\n\n// Delete specific document chunks by source\napp.delete('/api/documents/:source', (req, res) => {\n const source = decodeURIComponent(req.params.source);\n const beforeCount = documentChunks.length;\n documentChunks = documentChunks.filter(chunk => chunk.source !== source);\n const removed = beforeCount - documentChunks.length;\n \n res.json({\n message: `Removed ${removed} chunks from ${source}`,\n remainingChunks: documentChunks.length\n });\n});\n\n// Clear all documents\napp.delete('/api/documents', (req, res) => {\n const count = documentChunks.length;\n documentChunks = [];\n res.json({\n message: `Cleared ${count} chunks`,\n remainingChunks: 0\n });\n});\n\n// Endpoint to chat using RAG (with optional streaming)\napp.post('/api/chat', async (req, res) => {\n console.log(`[ROUTE] Incoming POST /api/chat - Body Keys: ${Object.keys(req.body || {}).join(', ')}`);\n try {\n const { \n messages, \n personality, \n mode, \n jurisdiction, \n basePrompt, \n stream = false,\n provider = 'auto', \n apiKeys = {} \n } = req.body;\n \n if (!messages || !Array.isArray(messages) || messages.length === 0) {\n return res.status(400).json({ error: 'Messages array is required' });\n }\n\n // Get the latest user message to search for context\n const latestUserMessage = messages[messages.length - 1].content;\n \n let contextStr = '';\n let retrievedSources = [];\n \n if (documentChunks.length > 0) {\n try {\n const userVector = await embeddings.embedQuery(latestUserMessage);\n \n // Calculate scores for all chunks, prioritizing ones that match the jurisdiction if specified\n const scoredChunks = documentChunks.map(chunk => {\n let score = cosineSimilarity(userVector, chunk.vector);\n \n // Boost chunks from the correct jurisdiction (filename-based tagging)\n if (jurisdiction && jurisdiction !== 'National' && chunk.source?.toLowerCase().includes(jurisdiction.toLowerCase())) {\n score += 0.2;\n }\n \n return {\n content: chunk.content,\n score: score,\n source: chunk.source\n };\n });\n \n // Sort descending by score\n scoredChunks.sort((a, b) => b.score - a.score);\n \n // Take top 5 chunks for richer context\n const topChunks = scoredChunks.slice(0, 5).filter(c => c.score > 0.1); // Filter low-relevance chunks\n \n retrievedSources = [...new Set(topChunks.map(c => c.source))];\n \n contextStr = \"\\n\\nCRITICAL LEGAL PRECEDENTS (From Knowledge Base):\\n\" + \n topChunks.map(c => `[Source: ${c.source}] [Relevance: ${(c.score * 100).toFixed(1)}%]\\n${c.content}`).join(\"\\n\\n---\\n\\n\");\n \n console.log(` 🔍 RAG retrieved top chunks from: ${retrievedSources.join(', ')}`);\n } catch (embedError) {\n console.warn('Embedding failed, proceeding without RAG context:', embedError.message);\n }\n }\n\n // Format the prompt\n let systemPrompt = basePrompt || `You are JusticeAI, a powerful legal co-pilot specializing in Indian law.`;\n \n systemPrompt += `\\n\\nCURRENT CONTEXT: You are providing assistance under the jurisdiction of ${jurisdiction || 'India (National)'}.`;\n\n if (mode === 'simulator') {\n systemPrompt += `\\n\\nROLE PLAY: You are opposing counsel. Be adversarial and challenge the user's arguments aggressively. Exploit weaknesses in their position. This is a simulation for moot practice only.`;\n }\n if (personality) {\n systemPrompt += `\\n\\nYour judicial/opponent personality is: ${personality}.`;\n }\n \n if (mode === 'analyzer') {\n systemPrompt += `\\n\\nCRITICAL RULE: You must return your response strictly as a structured JSON object.`;\n } else {\n systemPrompt += `\\n\\nCRITICAL CONVERSATIONAL RULE: Keep your conversational responses clear, articulate, and well-formatted. Do not output raw JSON objects in this mode. Use elegant bullet points and line breaks.`;\n }\n\n systemPrompt += contextStr;\n\n // Build the request for Ollama\n const requestData = {\n model: CHAT_MODEL,\n messages: [\n { role: 'system', content: systemPrompt },\n ...messages.map(m => ({\n role: m.role,\n content: m.content\n }))\n ],\n stream: stream\n };\n\n console.log(`Routing query to Ollama: \"${latestUserMessage.substring(0, 50)}...\"`);\n \n try {\n if (stream) {\n // Streaming response\n res.setHeader('Content-Type', 'text/event-stream');\n res.setHeader('Cache-Control', 'no-cache');\n res.setHeader('Connection', 'keep-alive');\n\n const response = await fetch(`${OLLAMA_BASE_URL}/api/chat`, {\n method: 'POST',\n headers: { 'Content-Type': 'application/json' },\n body: JSON.stringify(requestData),\n signal: AbortSignal.timeout(10000) // 10s timeout for local Ollama\n });\n\n if (!response.ok) {\n throw new Error(`Ollama Error: ${response.statusText}`);\n }\n\n // Pipe the streaming response\n const reader = response.body.getReader();\n const decoder = new TextDecoder();\n\n while (true) {\n const { done, value } = await reader.read();\n if (done) break;\n \n const chunk = decoder.decode(value);\n res.write(`data: ${chunk}\\n\\n`);\n }\n\n res.end();\n } else {\n // Non-streaming response\n const ollamaResponse = await fetch(`${OLLAMA_BASE_URL}/api/chat`, {\n method: 'POST',\n headers: { 'Content-Type': 'application/json' },\n body: JSON.stringify(requestData),\n signal: AbortSignal.timeout(15000) // 15s timeout\n });\n\n if (!ollamaResponse.ok) {\n throw new Error(`Ollama Error: ${await ollamaResponse.text()}`);\n }\n\n const data = await ollamaResponse.json();\n res.json({ \n message: { \n role: 'assistant', \n content: data.message.content \n },\n metadata: {\n sources: retrievedSources,\n model: CHAT_MODEL,\n provider: 'Ollama (Local)',\n timestamp: new Date().toISOString()\n }\n });\n }\n } catch (ollamaErr) {\n console.warn(`⚠️ Ollama failure detected: ${ollamaErr.message}. Shifting to Gemini fallback...`);\n \n let finalResult = null;\n let usedProvider = null;\n let errorChain = [`Ollama: ${ollamaErr.message}`];\n\n if (provider !== 'auto') {\n console.log(`User-forced provider: ${provider}`);\n if (provider === 'gemini') {\n finalResult = await callGemini(messages, systemPrompt, apiKeys.gemini);\n usedProvider = 'Gemini (User Controlled)';\n } else if (provider === 'deepseek') {\n finalResult = await callDeepSeek(messages, systemPrompt, apiKeys.deepseek);\n usedProvider = 'DeepSeek (User Controlled)';\n } else {\n // Fall back to Ollama if forced provider is ollama or invalid\n throw ollamaErr; \n }\n } else {\n // Waterfall Tier 2: Gemini\n if (apiKeys.gemini || GEMINI_API_KEY) {\n try {\n finalResult = await callGemini(messages, systemPrompt, apiKeys.gemini);\n usedProvider = 'Gemini (Cloud Fallback)';\n } catch (geminiErr) {\n console.warn(`⚠️ Gemini fallback failed: ${geminiErr.message}. Shifting to DeepSeek...`);\n errorChain.push(`Gemini: ${geminiErr.message}`);\n }\n }\n\n // Waterfall Tier 3: DeepSeek\n if (!finalResult && (apiKeys.deepseek || DEEPSEEK_API_KEY)) {\n try {\n finalResult = await callDeepSeek(messages, systemPrompt, apiKeys.deepseek);\n usedProvider = 'DeepSeek (Cloud Fallback)';\n } catch (deepseekErr) {\n console.error(`❌ DeepSeek fallback failed: ${deepseekErr.message}`);\n errorChain.push(`DeepSeek: ${deepseekErr.message}`);\n }\n }\n }\n\n if (finalResult) {\n res.json({\n ...finalResult,\n metadata: {\n sources: retrievedSources,\n provider: usedProvider,\n timestamp: new Date().toISOString(),\n errorChain: errorChain\n }\n });\n } else {\n throw new Error(`Multi-tier failover exhausted. Chain: ${errorChain.join(' -> ')}`);\n }\n }\n \n } catch (err) {\n console.error(\"Chat Error:\", err);\n res.status(500).json({ \n error: err.message,\n details: process.env.NODE_ENV === 'development' ? err.stack : undefined\n });\n }\n});\n\n// Embedding endpoint (for testing/debugging)\napp.post('/api/embed', async (req, res) => {\n try {\n const { text } = req.body;\n if (!text) {\n return res.status(400).json({ error: 'Text is required' });\n }\n \n const vector = await embeddings.embedQuery(text);\n res.json({\n vector: vector.slice(0, 10), // Return first 10 dimensions for brevity\n dimensions: vector.length,\n model: EMBEDDING_MODEL\n });\n } catch (err) {\n res.status(500).json({ error: err.message });\n }\n});\n\n// Error handling middleware\napp.use((err, req, res, next) => {\n console.error('Unhandled error:', err);\n res.status(500).json({\n error: 'Internal server error',\n message: process.env.NODE_ENV === 'development' ? err.message : 'Something went wrong'\n });\n});\n\n// 404 handler\napp.use((req, res) => {\n res.status(404).json({\n error: 'Not found',\n path: req.path,\n method: req.method\n });\n});\n\n// Start the server\n// Relocating document indexing to keep the server alive correctly\n(async () => {\n await loadAndIndexDocuments();\n})();\n\n// Keep process alive if event loop becomes empty (Express should handle this, but adding a fail-safe)\nsetInterval(() => {}, 1000 * 60 * 60); // 1 hour tick\n\napp.listen(PORT, () => {\n console.log(`\\n⚖️ JusticeAI RAG Server running on port ${PORT}`);\n console.log(`📡 API available at: http://localhost:${PORT}/api`);\n console.log(`🏥 Health check: http://localhost:${PORT}/api/health`);\n console.log(`\\nMake sure Ollama is running:`);\n console.log(` ollama serve`);\n console.log(`\\nRequired models:`);\n console.log(` ollama pull ${EMBEDDING_MODEL}`);\n console.log(` ollama pull ${CHAT_MODEL}\\n`);\n});\n\n// Graceful shutdown\nprocess.on('SIGTERM', () => {\n console.log('SIGTERM received, shutting down gracefully...');\n process.exit(0);\n});\n\nprocess.on('SIGINT', () => {\n console.log('SIGINT received, shutting down gracefully...');\n process.exit(0);\n});"}}}}