Skip to Content
Last modified by Robert Schaub on 2026/02/08 08:12
From version 2.1
edited by Robert Schaub
on 2026/01/02 10:01
Change comment: There is no comment for this version
To version 3.1
edited by Robert Schaub
on 2026/01/02 10:02
Change comment: There is no comment for this version

Summary

Details

Page properties
Content
... ... @@ -5,8 +5,6 @@
5 5  **Analysis Date:** January 2026
6 6  **Document Purpose:** Technical diagrams, gap analysis, and optimization recommendations
7 7  
8 ------
9 -
10 10  ----
11 11  
12 12  == 1. AKEL Flow Diagram (with LLM and WebSearch Interactions) ==
... ... @@ -93,7 +93,7 @@
93 93   class UNDERSTAND,DECIDE,FETCHSRC,EXTRACT,VERDICT,REPORT step
94 94  {{/mermaid}}
95 95  
96 ------
94 +----
97 97  
98 98  
99 99  == 2. ERD Data Model (Current POC1 Implementation) ==
... ... @@ -188,7 +188,7 @@
188 188   }
189 189  {{/mermaid}}
190 190  
191 ------
189 +----
192 192  
193 193  
194 194  == 3. Overall Architecture with Interactions ==
... ... @@ -287,7 +287,7 @@
287 287   class ANALYZE_API,JOBS_API,JOB_API,EVENTS_API,RUN_JOB api
288 288  {{/mermaid}}
289 289  
290 ------
288 +----
291 291  
292 292  
293 293  == 4. Specification vs Implementation Gap Analysis ==
... ... @@ -326,7 +326,7 @@
326 326  
327 327  
328 328  | Spec Requirement | POC1 Status | Gap Description |
329 -||-|-|
327 +| |-|-|
330 330  | **Three-Layer Architecture** | ✅ Implemented | Interface (Next.js) → Processing (AKEL) → Data (SQLite) |
331 331  | **LLM Abstraction Layer** | ✅ Implemented | AI SDK supports multiple providers with failover |
332 332  | **PostgreSQL Primary DB** | ⚠️ Different | Using SQLite for simplicity (acceptable for POC) |
... ... @@ -339,13 +339,13 @@
339 339  
340 340  
341 341  | Spec Feature | POC1 Status | Gap Description |
342 -||-|-|
340 +| |-|-|
343 343  | **Risk Tier Publication Rules** | ❌ Missing | All results published immediately regardless of tier |
344 344  | **Human Review Queue** | ❌ Missing | No review workflow |
345 345  | **AI-Generated Labeling** | ⚠️ Partial | Results show "AI analysis" but no formal labeling system |
346 346  | **Audit Rate Sampling** | ❌ Missing | No sampling audits |
347 347  
348 ------
346 +----
349 349  
350 350  
351 351  == 5. Optimization Recommendations ==
... ... @@ -363,7 +363,7 @@
363 363  {{/mermaid}}
364 364  
365 365  | Optimization | Estimated Savings | Implementation Effort |
366 -||-----||
364 +| |-| |
367 367  | **Cache claim understanding** | 30-50% on repeated claims | Medium |
368 368  | **Use Haiku for fact extraction** | 40% on Step 2 costs | Low (config change) |
369 369  | **Batch fact extraction** | 20% fewer API calls | Medium |
... ... @@ -399,7 +399,7 @@
399 399  {{/mermaid}}
400 400  
401 401  | Optimization | Time Savings | Notes |
402 -|||-----|
400 +| | |-|
403 403  | **Parallel source fetching** | Already implemented | Currently fetches 3 sources in parallel |
404 404  | **Streaming LLM responses** | 20-30% perceived | User sees progress faster |
405 405  | **Search query batching** | 10-15% | Send multiple queries to search API |
... ... @@ -424,7 +424,7 @@
424 424   - Cache search results (1h TTL)
425 425   - Reduces external API calls
426 426  
427 ------
425 +----
428 428  
429 429  
430 430  == 6. Separated Verdict Architecture Proposal ==
... ... @@ -510,7 +510,7 @@
510 510  
511 511  
512 512  | Benefit | Impact | Rationale |
513 -|-| |-----|
511 +|-| |-|
514 514  | **Cost Reduction** | 40-70% for repeated claims | Many articles share common claims (e.g., "COVID vaccines are safe") |
515 515  | **Faster Analysis** | 50%+ for cached claims | Skip research + LLM calls for known claims |
516 516  | **Consistency** | High | Same claim always gets same verdict (until cache expires) |
... ... @@ -554,7 +554,7 @@
554 554   - Phase 2: Semantic similarity caching (embedding-based)
555 555   - Phase 3: Federated claim sharing across instances
556 556  
557 ------
555 +----
558 558  
559 559  
560 560  == 7. Summary ==