Spaces:
Sleeping
Sleeping
Srini P commited on
Commit Β·
c28eaa9
1
Parent(s): 3d20163
Docs: Update system documentation with latest model info and recent feature improvements
Browse files- COMPLETE_SYSTEM_GUIDE.md +1 -1
- README.md +6 -6
- app/backend/ARCHITECTURE.md +1 -1
COMPLETE_SYSTEM_GUIDE.md
CHANGED
|
@@ -32,7 +32,7 @@ This document provides an overview of the complete FinBot RAG system with both f
|
|
| 32 |
β 1. Input Guards (injection, PII, etc) β
|
| 33 |
β 2. Query Router (semantic routing) β
|
| 34 |
β 3. RBAC Retriever (metadata filtering) β
|
| 35 |
-
β 4. LLM Generation (
|
| 36 |
β 5. Output Guards (grounding, citations) β
|
| 37 |
ββββββββββββββββββ¬ββββββββββββββββββββββββββ
|
| 38 |
β
|
|
|
|
| 32 |
β 1. Input Guards (injection, PII, etc) β
|
| 33 |
β 2. Query Router (semantic routing) β
|
| 34 |
β 3. RBAC Retriever (metadata filtering) β
|
| 35 |
+
β 4. LLM Generation (Llama 3.3 70B) β
|
| 36 |
β 5. Output Guards (grounding, citations) β
|
| 37 |
ββββββββββββββββββ¬ββββββββββββββββββββββββββ
|
| 38 |
β
|
README.md
CHANGED
|
@@ -65,11 +65,11 @@ FinBot solves both problems:
|
|
| 65 |
βββββββββββββ ββββββββββββββββ ββββββββββββββ ββββββββββββ
|
| 66 |
β GUARDRAILS β β SEMANTIC β β RBAC β β LLM β
|
| 67 |
β (INPUT) β β ROUTING β βRETRIEVAL β β (GROQ) β
|
| 68 |
-
β β β β β β β
|
| 69 |
-
β β’ Injection β β β’ 5 Routes β β β’ Filter β β
|
| 70 |
-
β β’ Off-topic β β β’ Collectionsβ β β’ By Role β β β’
|
| 71 |
-
β β’ PII β β β’ Role Check β β β’ Qdrant β β β’
|
| 72 |
-
β β’ Rate limit β β β β β β
|
| 73 |
βββββββββββββ ββββββββββββββββ ββββββββββββββ ββββββββββββ
|
| 74 |
β β
|
| 75 |
ββββββββββ¬βββββββββ
|
|
@@ -214,7 +214,7 @@ Assignment1/
|
|
| 214 |
### Prerequisites
|
| 215 |
|
| 216 |
- Python 3.10+
|
| 217 |
-
-
|
| 218 |
- ~500MB disk space for Qdrant
|
| 219 |
- Modern web browser
|
| 220 |
|
|
|
|
| 65 |
βββββββββββββ ββββββββββββββββ ββββββββββββββ ββββββββββββ
|
| 66 |
β GUARDRAILS β β SEMANTIC β β RBAC β β LLM β
|
| 67 |
β (INPUT) β β ROUTING β βRETRIEVAL β β (GROQ) β
|
| 68 |
+
β β β β β β β(Llama 3.3β
|
| 69 |
+
β β’ Injection β β β’ 5 Routes β β β’ Filter β β 70B) β
|
| 70 |
+
β β’ Off-topic β β β’ Collectionsβ β β’ By Role β β β’ Answer β
|
| 71 |
+
β β’ PII β β β’ Role Check β β β’ Qdrant β β β’ Cite β
|
| 72 |
+
β β’ Rate limit β β β β β β β’ Ground β
|
| 73 |
βββββββββββββ ββββββββββββββββ ββββββββββββββ ββββββββββββ
|
| 74 |
β β
|
| 75 |
ββββββββββ¬βββββββββ
|
|
|
|
| 214 |
### Prerequisites
|
| 215 |
|
| 216 |
- Python 3.10+
|
| 217 |
+
- Groq API key
|
| 218 |
- ~500MB disk space for Qdrant
|
| 219 |
- Modern web browser
|
| 220 |
|
app/backend/ARCHITECTURE.md
CHANGED
|
@@ -129,7 +129,7 @@ User Role + Selected Collection β Check access β Query vector store
|
|
| 129 |
|
| 130 |
**Stage 4: LLM Generation**
|
| 131 |
```
|
| 132 |
-
Question + Retrieved Chunks β Groq (
|
| 133 |
Uses retrieved chunks as context (RAG)
|
| 134 |
```
|
| 135 |
|
|
|
|
| 129 |
|
| 130 |
**Stage 4: LLM Generation**
|
| 131 |
```
|
| 132 |
+
Question + Retrieved Chunks β Groq (Llama 3.3 70B) β Generate answer
|
| 133 |
Uses retrieved chunks as context (RAG)
|
| 134 |
```
|
| 135 |
|