Skip to Content
Last modified by Robert Schaub on 2025/12/24 20:16
Hide last authors
Robert Schaub 1.1 1 {{mermaid}}
2 graph LR
3 subgraph AKEL["AKEL Pipeline"]
4 S1[Stage 1<br/>Extract Claims]
5 S2[Stage 2<br/>Analyze Claims]
6 S3[Stage 3<br/>Holistic Assessment]
7 end
8
9 subgraph LLM["LLM Abstraction Layer"]
10 INT[Provider Interface]
11 CFG[Configuration<br/>Registry]
12 FAIL[Failover<br/>Handler]
13 end
14
15 subgraph Providers["LLM Providers"]
16 ANT[Anthropic<br/>Claude API<br/>PRIMARY]
17 OAI[OpenAI<br/>GPT API<br/>SECONDARY]
18 GOO[Google<br/>Gemini API<br/>TERTIARY]
19 LOC[Local Models<br/>Llama/Mistral<br/>FUTURE]
20 end
21
22 S1 --> INT
23 S2 --> INT
24 S3 --> INT
25
26 INT --> CFG
27 INT --> FAIL
28
29 CFG --> ANT
30 FAIL --> ANT
31 FAIL --> OAI
32 FAIL --> GOO
33
34 ANT -.fallback.-> OAI
35 OAI -.fallback.-> GOO
36
37 style AKEL fill:#ffcccc
38 style LLM fill:#ccffcc
39 style Providers fill:#e1f5ff
40 style ANT fill:#ff9999
41 style OAI fill:#99ccff
42 style GOO fill:#99ff99
43 style LOC fill:#cccccc
44 {{/mermaid}}
45
Robert Schaub 2.1 46 **LLM Abstraction Architecture** - AKEL stages call through provider interface. Configuration registry selects provider per stage. Failover handler implements automatic fallback chain.
47
48 **POC1 Implementation:**
49 * PRIMARY: Anthropic Claude API (Haiku 4.5 for Stage 1, Sonnet 4.5 for Stages 2 & 3)
50 * Failover: Basic error handling with cache fallback
51
52 **Future (POC2/Beta):**
53 * SECONDARY: OpenAI GPT API (automatic failover)
54 * TERTIARY: Google Gemini API (tertiary fallback)
55 * FUTURE: Local models (Llama/Mistral for on-premises deployments)
56
57 **Architecture Benefits:**
58 * Prevents vendor lock-in
59 * Ensures resilience through automatic failover
60 * Enables cost optimization per stage
61 * Supports regulatory compliance (provider selection for data residency)
62