RAG Performance
Retrieval quality, knowledge gap analysis, per-document hit rates, thumbs feedback scores, and LLM token cost breakdown.
RAG Queries (30d)
8,421
Avg Confidence
0.79
โ 0.04
Below Threshold
12%
threshold: 0.45
Thumbs Up Rate
86%
Total Token Cost
โน4,821
โ 8% MoM
๐ Confidence Score Distribution
0โ0.2
0.2โ0.4
0.4โ0.5
0.5โ0.6
0.6โ0.7
0.7โ0.8
0.8โ0.9
0.9โ1.0
โ Below threshold (12%)
โ Borderline (14%)
โ High confidence (74%)
๐ Per-KB Hit Rate & Avg Confidence
Admissions & Enrollment
2,847 hits0.84
Fee Structure 2025-26
1,923 hits0.87
Academic Calendar
1,241 hits0.81
Campus Life & Facilities
892 hits0.68
Examination Rules
487 hits0.72
๐ Top Retrieved Documents
| Document | Retrievals | Avg Rank | Avg Sim | Citations Used |
|---|---|---|---|---|
Fee Structure 2025-26 PDF ยท 42 chunks | 1,923 | 1.4 | 0.87 | 1,247 |
Admissions Brochure 2025 PDF ยท 87 chunks | 1,641 | 1.7 | 0.84 | 1,102 |
Academic Calendar 2025-26 PDF ยท 28 chunks | 1,241 | 1.2 | 0.82 | 987 |
Hostel Rules 2025 PDF ยท 31 chunks | 487 | 2.1 | 0.74 | 312 |
Scholarship FAQ Manual Q&A ยท 24 chunks | 341 | 1.9 | 0.71 | 198 |
๐๐ Feedback Review Queue
1,247
๐ Positive
198
๐ Negative
47
โณ Unreviewed
๐ Negative feedback
"What is the process for PhD admission?" โ Bot answer was incomplete, missed eligibility criteria.
๐ Negative feedback
"Are there NRI seats available?" โ Bot couldn't find relevant information (confidence 0.22).
๐ฐ Token & Cost Breakdown (30d)
24.7M
Total tokens
18.2M
Input (prompt)
6.5M
Output (completion)
โน4,821
Total cost ยท โน160/day avg
| Model | Queries | Avg tokens/query | Input tokens | Output tokens | Cost | Cost/query |
|---|---|---|---|---|---|---|
| gpt-4o Primary | 5,842 | 2,847 | 14.2M | 4.8M | โน3,941 | โน0.67 |
| gpt-4o-mini Cheap tier | 2,341 | 1,823 | 3.8M | 1.5M | โน724 | โน0.31 |
| text-embedding-3-small Embeddings | 8,421 | 284 | 2.4M | โ | โน156 | โน0.02 |