import { Hono } from 'hono'; import { streamSSE } from 'hono/streaming'; import { v4 as uuidv4 } from 'uuid'; import { z } from 'zod'; import { config } from '../config'; import { cache } from '../services/cache'; import { vespaRequest } from '../services/vespa-https'; const api = new Hono(); // Search request schema const searchQuerySchema = z.object({ query: z.string().min(1).max(500), ranking: z.enum(['hybrid', 'colpali', 'bm25']).optional().default('hybrid'), }); // Main search endpoint api.get('/colpali-search', async (c) => { try { const query = c.req.query('query'); const ranking = c.req.query('ranking') || 'hybrid'; const validation = searchQuerySchema.safeParse({ query, ranking }); if (!validation.success) { return c.json({ error: 'Invalid request', details: validation.error.issues }, 400); } const validatedData = validation.data; // Check cache const cacheKey = `search:${validatedData.query}:${validatedData.ranking}`; const cachedResult = cache.get(cacheKey); if (cachedResult) { c.header('X-Cache', 'HIT'); return c.json(cachedResult); } // Build YQL query based on ranking let yql = ''; let rankProfile = 'default'; switch (validatedData.ranking) { case 'colpali': yql = `select * from linqto where userQuery() limit 20`; rankProfile = 'colpali'; break; case 'bm25': yql = `select * from linqto where userQuery() limit 20`; rankProfile = 'bm25'; break; case 'hybrid': default: yql = `select * from linqto where userQuery() limit 20`; rankProfile = 'default'; break; } // Query Vespa directly const searchUrl = `${config.vespaAppUrl}/search/`; const searchParams = new URLSearchParams({ yql, query: validatedData.query, ranking: rankProfile, hits: '20' }); const response = await vespaRequest(`${searchUrl}?${searchParams}`); if (!response.ok) { const errorText = await response.text(); console.error('Vespa error:', errorText); throw new Error(`Vespa returned ${response.status}: ${errorText}`); } const data = await response.json(); // Generate query_id for sim_map compatibility const queryId = uuidv4(); // Transform to match expected format if (data.root && data.root.children) { data.root.children.forEach((hit: any, idx: number) => { if (!hit.fields) hit.fields = {}; // Add sim_map identifier for compatibility hit.fields.sim_map = `${queryId}_${idx}`; }); } // Cache the result cache.set(cacheKey, data); c.header('X-Cache', 'MISS'); return c.json(data); } catch (error) { console.error('Search error:', error); return c.json({ error: 'Search failed', message: error instanceof Error ? error.message : 'Unknown error' }, 500); } }); // Full image endpoint api.get('/full-image', async (c) => { try { const docId = c.req.query('docId'); if (!docId) { return c.json({ error: 'docId is required' }, 400); } // Check cache const cacheKey = `fullimage:${docId}`; const cachedImage = cache.get<{ base64_image: string }>(cacheKey); if (cachedImage) { c.header('X-Cache', 'HIT'); return c.json(cachedImage); } // Query Vespa for the document const searchUrl = `${config.vespaAppUrl}/search/`; const searchParams = new URLSearchParams({ yql: `select * from linqto where id contains "${docId}"`, hits: '1' }); const response = await vespaRequest(`${searchUrl}?${searchParams}`); if (!response.ok) { throw new Error(`Vespa returned ${response.status}`); } const data = await response.json(); if (data.root?.children?.[0]?.fields) { const fields = data.root.children[0].fields; const base64Image = fields.full_image || fields.image; if (base64Image) { const result = { base64_image: base64Image }; cache.set(cacheKey, result, 86400); // 24 hours c.header('X-Cache', 'MISS'); return c.json(result); } } return c.json({ error: 'Image not found' }, 404); } catch (error) { console.error('Full image error:', error); return c.json({ error: 'Failed to fetch image', message: error instanceof Error ? error.message : 'Unknown error' }, 500); } }); // Query suggestions endpoint api.get('/query-suggestions', async (c) => { try { const query = c.req.query('query'); // Static suggestions for now const staticSuggestions = [ 'linqto bankruptcy', 'linqto filing date', 'linqto creditors', 'linqto assets', 'linqto liabilities', 'linqto chapter 11', 'linqto docket', 'linqto plan', 'linqto disclosure statement', 'linqto claims', ]; if (!query) { return c.json({ suggestions: staticSuggestions.slice(0, 5) }); } const lowerQuery = query.toLowerCase(); const filtered = staticSuggestions .filter(s => s.toLowerCase().includes(lowerQuery)) .slice(0, 5); return c.json({ suggestions: filtered }); } catch (error) { console.error('Suggestions error:', error); return c.json({ error: 'Failed to fetch suggestions', suggestions: [] }, 500); } }); // Similarity maps endpoint (placeholder) api.get('/similarity-maps', async (c) => { try { const queryId = c.req.query('queryId'); const idx = c.req.query('idx'); const token = c.req.query('token'); const tokenIdx = c.req.query('tokenIdx'); if (!queryId || !idx || !token || !tokenIdx) { return c.json({ error: 'Missing required parameters' }, 400); } // Return placeholder HTML const html = `

Similarity Map

Query: ${token}

Document: ${idx}

Similarity map generation requires the ColPali model. This is a placeholder for the demo.

`; return c.html(html); } catch (error) { console.error('Similarity map error:', error); return c.json({ error: 'Failed to generate similarity map', message: error instanceof Error ? error.message : 'Unknown error' }, 500); } }); // Visual RAG Chat SSE endpoint api.get('/visual-rag-chat', async (c) => { const queryId = c.req.query('queryId'); const query = c.req.query('query'); const docIds = c.req.query('docIds'); if (!queryId || !query || !docIds) { return c.json({ error: 'Missing required parameters: queryId, query, docIds' }, 400); } return streamSSE(c, async (stream) => { try { // Mock response for now - in production this would use an LLM const messages = [ `I'll analyze the search results for your query: "${query}"`, "Based on the documents provided, here are the key findings:", "1. LINQTO filed for Chapter 11 bankruptcy protection", "2. The filing includes detailed financial statements and creditor information", "3. Various claims and assets are documented in the court filings", "", "This is a demo response. In production, this would analyze the actual document contents using an LLM." ]; for (const msg of messages) { await stream.writeSSE({ data: msg }); await new Promise(resolve => setTimeout(resolve, 300)); // Simulate typing } } catch (error) { console.error('Chat streaming error:', error); await stream.writeSSE({ event: 'error', data: JSON.stringify({ error: 'Streaming failed', message: error instanceof Error ? error.message : 'Unknown error' }), }); } }); }); export { api };