Notes

(.abhi2) [pfls@ip-172-28-12-89 output]$ # On the VM — find the latest report
cat $(ls -t /home/pfls/abhishek/src/output/latency_report_*.txt 2>/dev/null || ls -t output/latency_report_*.txt | head -1)
======================================================================
CREDIT-RISK LATENCY ANALYSIS & BENCHMARK REPORT: 8e8e2725-cb98-4796-bbfa-cc11b532a990
======================================================================
Process PID : 323792
Session Start : 2026-06-02T12:38:03.174913
End-to-End Latency : 160.37 seconds
Request Status : SUCCESS
Total Documents Processed : 15
----------------------------------------------------------------------
PIPELINE STAGE LATENCY SUMMARY:
- S3 / Local Download : 2.54s ( 1.6% of total time)
- Azure OCR Engine : 39.48s ( 24.6% of total time)
- Embedding Generation : 5.75s ( 3.6% of total time)
- Vector Retrieval : 4.02s ( 2.5% of total time)
- LLM Inference : 44.44s ( 27.7% of total time)

OTHER Global Orchestration Stages:
- Verification Pipeline Time Seconds: 0.03s
- Cibil Processing Time Seconds: 0.03s
- Final Pdf Generation Time Seconds: 0.20s
- S3 Upload Time Seconds : 0.09s
- Lsq Submission Time Seconds : 4.13s
----------------------------------------------------------------------
DETAILED PER-DOCUMENT TIMING SUMMARY:
* CA1/CIBIL (CIBIL.json):
- S3 Download : 0.11s
- Azure OCR : 0.00s (Pages: 0, Avg/Page: 0.00s)
- Embedding Gen : 0.00s (Chunks: 0)
- Retrieval Time : 0.00s
- LLM Inference : 0.00s (Load: 0.00s, Prompt Eval: 0.00s)
(Tokens: In=0, Out=0, speed=0.0 tok/s)
- JSON parsing : 0.00s
- Total Doc Time : 0.17s

* Details/Details (Activity.json):
- S3 Download : 0.10s
- Azure OCR : 0.00s (Pages: 0, Avg/Page: 0.00s)
- Embedding Gen : 0.00s (Chunks: 0)
- Retrieval Time : 0.00s
- LLM Inference : 0.00s (Load: 0.00s, Prompt Eval: 0.00s)
(Tokens: In=0, Out=0, speed=0.0 tok/s)
- JSON parsing : 0.00s
- Total Doc Time : 0.16s

* Details/Details (Lead.json):
- S3 Download : 0.04s
- Azure OCR : 0.00s (Pages: 0, Avg/Page: 0.00s)
- Embedding Gen : 0.00s (Chunks: 0)
- Retrieval Time : 0.00s
- LLM Inference : 0.00s (Load: 0.00s, Prompt Eval: 0.00s)
(Tokens: In=0, Out=0, speed=0.0 tok/s)
- JSON parsing : 0.00s
- Total Doc Time : 0.04s

* KP/CIBIL (CIBIL.json):
- S3 Download : 0.10s
- Azure OCR : 0.00s (Pages: 0, Avg/Page: 0.00s)
- Embedding Gen : 0.00s (Chunks: 0)
- Retrieval Time : 0.00s
- LLM Inference : 0.00s (Load: 0.00s, Prompt Eval: 0.00s)
(Tokens: In=0, Out=0, speed=0.0 tok/s)
- JSON parsing : 0.00s
- Total Doc Time : 0.15s

* MA/CIBIL (CIBIL.json):
- S3 Download : 0.09s
- Azure OCR : 0.00s (Pages: 0, Avg/Page: 0.00s)
- Embedding Gen : 0.00s (Chunks: 0)
- Retrieval Time : 0.00s
- LLM Inference : 0.00s (Load: 0.00s, Prompt Eval: 0.00s)
(Tokens: In=0, Out=0, speed=0.0 tok/s)
- JSON parsing : 0.00s
- Total Doc Time : 0.13s

* CA1/Address (Business Address Proof [1].pdf):
- S3 Download : 0.21s
- Azure OCR : 5.52s (Pages: 3, Avg/Page: 1.84s)
- Embedding Gen : 2.74s (Chunks: 4)
- Retrieval Time : 0.00s
- LLM Inference : 1.17s (Load: 0.00s, Prompt Eval: 1.13s)
(TTFT: 0.043s, GPU Util: 100.0%, VRAM: 20211.0/23028.0 MiB)
(Tokens: In=502, Out=25, speed=22.1 tok/s)
- JSON parsing : 0.00s
- Total Doc Time : 9.68s

* CA1/UDYAM certificate (UDYAM certificate [1].pdf):
- S3 Download : 0.21s
- Azure OCR : 3.37s (Pages: 5, Avg/Page: 0.67s)
- Embedding Gen : 0.60s (Chunks: 14)
- Retrieval Time : 3.82s
- LLM Inference : 3.09s (Load: 0.00s, Prompt Eval: 3.04s)
(TTFT: 0.042s, GPU Util: 100.0%, VRAM: 20700.0/23028.0 MiB)
(Tokens: In=1309, Out=78, speed=25.6 tok/s)
- JSON parsing : 0.00s
- Total Doc Time : 11.13s

* CA1/GST Registration Certificate (GST Registration Certificate [1].pdf):
- S3 Download : 0.19s
- Azure OCR : 3.28s (Pages: 3, Avg/Page: 1.09s)
- Embedding Gen : 0.25s (Chunks: 4)
- Retrieval Time : 0.00s
- LLM Inference : 2.53s (Load: 0.00s, Prompt Eval: 2.49s)
(TTFT: 0.041s, GPU Util: 100.0%, VRAM: 20700.0/23028.0 MiB)
(Tokens: In=839, Out=68, speed=27.3 tok/s)
- JSON parsing : 0.00s
- Total Doc Time : 6.29s

* KP/Aadhar Card (Aadhar Card [1].pdf):
- S3 Download : 0.20s
- Azure OCR : 3.41s (Pages: 2, Avg/Page: 1.71s)
- Embedding Gen : 0.37s (Chunks: 1)
- Retrieval Time : 0.00s
- LLM Inference : 1.97s (Load: 0.00s, Prompt Eval: 1.93s)
(TTFT: 0.041s, GPU Util: 100.0%, VRAM: 20700.0/23028.0 MiB)
(Tokens: In=1119, Out=41, speed=21.3 tok/s)
- JSON parsing : 0.00s
- Total Doc Time : 5.99s

* KP/Address (Key Person POA - Aadhar [1].pdf):
- S3 Download : 0.20s
- Azure OCR : 3.24s (Pages: 2, Avg/Page: 1.62s)
- Embedding Gen : 0.19s (Chunks: 1)
- Retrieval Time : 0.00s
- LLM Inference : 0.91s (Load: 0.00s, Prompt Eval: 0.87s)
(TTFT: 0.036s, GPU Util: 100.0%, VRAM: 20700.0/23028.0 MiB)
(Tokens: In=242, Out=19, speed=21.7 tok/s)
- JSON parsing : 0.00s
- Total Doc Time : 4.57s

* KP/Address (Key Person POA - Aadhar [2].pdf):
- S3 Download : 0.18s
- Azure OCR : 3.24s (Pages: 2, Avg/Page: 1.62s)
- Embedding Gen : 0.48s (Chunks: 5)
- Retrieval Time : 0.00s
- LLM Inference : 1.18s (Load: 0.00s, Prompt Eval: 1.14s)
(TTFT: 0.039s, GPU Util: 100.0%, VRAM: 20700.0/23028.0 MiB)
(Tokens: In=567, Out=25, speed=22.0 tok/s)
- JSON parsing : 0.00s
- Total Doc Time : 5.11s

* KP/PAN Card (PAN Card [1].pdf):
- S3 Download : 0.22s
- Azure OCR : 3.29s (Pages: 2, Avg/Page: 1.64s)
- Embedding Gen : 0.23s (Chunks: 3)
- Retrieval Time : 0.00s
- LLM Inference : 1.55s (Load: 0.00s, Prompt Eval: 1.51s)
(TTFT: 0.041s, GPU Util: 100.0%, VRAM: 20700.0/23028.0 MiB)
(Tokens: In=962, Out=27, speed=17.9 tok/s)
- JSON parsing : 0.00s
- Total Doc Time : 5.33s

* MA/Address (Business Address Proof [1].pdf):
- S3 Download : 0.23s
- Azure OCR : 3.34s (Pages: 3, Avg/Page: 1.11s)
- Embedding Gen : 0.24s (Chunks: 4)
- Retrieval Time : 0.00s
- LLM Inference : 1.17s (Load: 0.00s, Prompt Eval: 1.13s)
(TTFT: 0.038s, GPU Util: 100.0%, VRAM: 20700.0/23028.0 MiB)
(Tokens: In=502, Out=25, speed=22.1 tok/s)
- JSON parsing : 0.00s
- Total Doc Time : 5.02s

* MA/UDYAM certificate (UDYAM certificate [1].pdf):
- S3 Download : 0.23s
- Azure OCR : 7.47s (Pages: 5, Avg/Page: 1.49s)
- Embedding Gen : 0.40s (Chunks: 14)
- Retrieval Time : 0.20s
- LLM Inference : 28.34s (Load: 0.00s, Prompt Eval: 28.30s)
(TTFT: 0.042s, GPU Util: 92.0%, VRAM: 20700.0/23028.0 MiB)
(Tokens: In=1309, Out=385, speed=13.6 tok/s)
- JSON parsing : 0.00s
- Total Doc Time : 93.41s

* MA/GST Registration Certificate (GST Registration Certificate [1].pdf):
- S3 Download : 0.23s
- Azure OCR : 3.32s (Pages: 3, Avg/Page: 1.11s)
- Embedding Gen : 0.24s (Chunks: 4)
- Retrieval Time : 0.00s
- LLM Inference : 2.53s (Load: 0.00s, Prompt Eval: 2.49s)
(TTFT: 0.041s, GPU Util: 100.0%, VRAM: 20700.0/23028.0 MiB)
(Tokens: In=839, Out=68, speed=27.3 tok/s)
- JSON parsing : 0.00s
- Total Doc Time : 6.36s

----------------------------------------------------------------------
BOTTLENECK ANALYSIS & MIGRATION RECOMMENDATIONS:
- Slowest Document : MA/UDYAM certificate (93.41s)
- Primary Bottleneck : LLM Inference (44.44s, 27.7%)

[RECOMMENDATION] Optimize LLM Inference Layer:
- If using Ollama, consider replacing it with vLLM which supports Continuous Batching and PagedAttention, increasing throughput by 3x-10x.
- If already using vLLM, optimize using tensor parallelism, model quantization (AWQ/GPTQ), or by adjusting VRAM constraints.
======================================================================
======================================================================
CREDIT-RISK LATENCY ANALYSIS & BENCHMARK REPORT: 1590863d-c7eb-4730-bb29-9adb934f2e0e
======================================================================
Process PID : 321173
Session Start : 2026-06-02T12:25:54.308650
End-to-End Latency : 163.71 seconds
Request Status : SUCCESS
Total Documents Processed : 15
----------------------------------------------------------------------
PIPELINE STAGE LATENCY SUMMARY:
- S3 / Local Download : 2.58s ( 1.6% of total time)
- Azure OCR Engine : 35.35s ( 21.6% of total time)
- Embedding Generation : 6.41s ( 3.9% of total time)
- Vector Retrieval : 4.07s ( 2.5% of total time)
- LLM Inference : 108.17s ( 66.1% of total time)

OTHER Global Orchestration Stages:
- Verification Pipeline Time Seconds: 0.03s
- Cibil Processing Time Seconds: 0.03s
- Final Pdf Generation Time Seconds: 0.20s
- S3 Upload Time Seconds : 0.10s
- Lsq Submission Time Seconds : 3.89s
----------------------------------------------------------------------
DETAILED PER-DOCUMENT TIMING SUMMARY:
* CA1/CIBIL (CIBIL.json):
- S3 Download : 0.11s
- Azure OCR : 0.00s (Pages: 0, Avg/Page: 0.00s)
- Embedding Gen : 0.00s (Chunks: 0)
- Retrieval Time : 0.00s
- LLM Inference : 0.00s (Load: 0.00s, Prompt Eval: 0.00s)
(Tokens: In=0, Out=0, speed=0.0 tok/s)
- JSON parsing : 0.00s
- Total Doc Time : 0.16s

* Details/Details (Activity.json):
- S3 Download : 0.10s
- Azure OCR : 0.00s (Pages: 0, Avg/Page: 0.00s)
- Embedding Gen : 0.00s (Chunks: 0)
- Retrieval Time : 0.00s
- LLM Inference : 0.00s (Load: 0.00s, Prompt Eval: 0.00s)
(Tokens: In=0, Out=0, speed=0.0 tok/s)
- JSON parsing : 0.00s
- Total Doc Time : 0.16s

* Details/Details (Lead.json):
- S3 Download : 0.04s
- Azure OCR : 0.00s (Pages: 0, Avg/Page: 0.00s)
- Embedding Gen : 0.00s (Chunks: 0)
- Retrieval Time : 0.00s
- LLM Inference : 0.00s (Load: 0.00s, Prompt Eval: 0.00s)
(Tokens: In=0, Out=0, speed=0.0 tok/s)
- JSON parsing : 0.00s
- Total Doc Time : 0.04s

* KP/CIBIL (CIBIL.json):
- S3 Download : 0.09s
- Azure OCR : 0.00s (Pages: 0, Avg/Page: 0.00s)
- Embedding Gen : 0.00s (Chunks: 0)
- Retrieval Time : 0.00s
- LLM Inference : 0.00s (Load: 0.00s, Prompt Eval: 0.00s)
(Tokens: In=0, Out=0, speed=0.0 tok/s)
- JSON parsing : 0.00s
- Total Doc Time : 0.13s

* MA/CIBIL (CIBIL.json):
- S3 Download : 0.12s
- Azure OCR : 0.00s (Pages: 0, Avg/Page: 0.00s)
- Embedding Gen : 0.00s (Chunks: 0)
- Retrieval Time : 0.00s
- LLM Inference : 0.00s (Load: 0.00s, Prompt Eval: 0.00s)
(Tokens: In=0, Out=0, speed=0.0 tok/s)
- JSON parsing : 0.00s
- Total Doc Time : 0.16s

* CA1/Address (Business Address Proof [1].pdf):
- S3 Download : 0.46s
- Azure OCR : 3.40s (Pages: 3, Avg/Page: 1.13s)
- Embedding Gen : 3.23s (Chunks: 4)
- Retrieval Time : 0.00s
- LLM Inference : 1.17s (Load: 0.00s, Prompt Eval: 1.13s)
(TTFT: 0.042s, GPU Util: 100.0%, VRAM: 20211.0/23028.0 MiB)
(Tokens: In=502, Out=25, speed=22.1 tok/s)
- JSON parsing : 0.00s
- Total Doc Time : 8.30s

* CA1/UDYAM certificate (UDYAM certificate [1].pdf):
- S3 Download : 0.20s
- Azure OCR : 3.38s (Pages: 5, Avg/Page: 0.68s)
- Embedding Gen : 0.78s (Chunks: 14)
- Retrieval Time : 3.86s
- LLM Inference : 47.48s (Load: 0.00s, Prompt Eval: 47.44s)
(TTFT: 0.042s, GPU Util: 87.0%, VRAM: 20700.0/23028.0 MiB)
(Tokens: In=1309, Out=591, speed=12.5 tok/s)
- JSON parsing : 0.00s
- Total Doc Time : 55.74s

* CA1/GST Registration Certificate (GST Registration Certificate [1].pdf):
- S3 Download : 0.18s
- Azure OCR : 3.34s (Pages: 3, Avg/Page: 1.11s)
- Embedding Gen : 0.24s (Chunks: 4)
- Retrieval Time : 0.00s
- LLM Inference : 2.53s (Load: 0.00s, Prompt Eval: 2.49s)
(TTFT: 0.041s, GPU Util: 100.0%, VRAM: 20700.0/23028.0 MiB)
(Tokens: In=839, Out=68, speed=27.3 tok/s)
- JSON parsing : 0.00s
- Total Doc Time : 6.32s

* KP/Aadhar Card (Aadhar Card [1].pdf):
- S3 Download : 0.16s
- Azure OCR : 3.29s (Pages: 2, Avg/Page: 1.65s)
- Embedding Gen : 0.38s (Chunks: 1)
- Retrieval Time : 0.00s
- LLM Inference : 1.97s (Load: 0.00s, Prompt Eval: 1.93s)
(TTFT: 0.041s, GPU Util: 100.0%, VRAM: 20700.0/23028.0 MiB)
(Tokens: In=1119, Out=41, speed=21.3 tok/s)
- JSON parsing : 0.00s
- Total Doc Time : 5.84s

* KP/Address (Key Person POA - Aadhar [1].pdf):
- S3 Download : 0.23s
- Azure OCR : 3.34s (Pages: 2, Avg/Page: 1.67s)
- Embedding Gen : 0.19s (Chunks: 1)
- Retrieval Time : 0.00s
- LLM Inference : 1.12s (Load: 0.00s, Prompt Eval: 1.08s)
(TTFT: 0.036s, GPU Util: 100.0%, VRAM: 20700.0/23028.0 MiB)
(Tokens: In=242, Out=25, speed=23.1 tok/s)
- JSON parsing : 0.00s
- Total Doc Time : 4.91s

* KP/Address (Key Person POA - Aadhar [2].pdf):
- S3 Download : 0.15s
- Azure OCR : 3.25s (Pages: 2, Avg/Page: 1.62s)
- Embedding Gen : 0.44s (Chunks: 5)
- Retrieval Time : 0.00s
- LLM Inference : 1.18s (Load: 0.00s, Prompt Eval: 1.14s)
(TTFT: 0.040s, GPU Util: 100.0%, VRAM: 20700.0/23028.0 MiB)
(Tokens: In=567, Out=25, speed=22.0 tok/s)
- JSON parsing : 0.00s
- Total Doc Time : 5.05s

* KP/PAN Card (PAN Card [1].pdf):
- S3 Download : 0.18s
- Azure OCR : 3.24s (Pages: 2, Avg/Page: 1.62s)
- Embedding Gen : 0.23s (Chunks: 3)
- Retrieval Time : 0.00s
- LLM Inference : 1.55s (Load: 0.00s, Prompt Eval: 1.51s)
(TTFT: 0.041s, GPU Util: 100.0%, VRAM: 20700.0/23028.0 MiB)
(Tokens: In=962, Out=27, speed=17.9 tok/s)
- JSON parsing : 0.00s
- Total Doc Time : 5.23s

* MA/Address (Business Address Proof [1].pdf):
- S3 Download : 0.19s
- Azure OCR : 3.30s (Pages: 3, Avg/Page: 1.10s)
- Embedding Gen : 0.26s (Chunks: 4)
- Retrieval Time : 0.00s
- LLM Inference : 1.17s (Load: 0.00s, Prompt Eval: 1.13s)
(TTFT: 0.039s, GPU Util: 100.0%, VRAM: 20700.0/23028.0 MiB)
(Tokens: In=502, Out=25, speed=22.1 tok/s)
- JSON parsing : 0.00s
- Total Doc Time : 4.96s

* MA/UDYAM certificate (UDYAM certificate [1].pdf):
- S3 Download : 0.19s
- Azure OCR : 4.47s (Pages: 5, Avg/Page: 0.89s)
- Embedding Gen : 0.42s (Chunks: 14)
- Retrieval Time : 0.21s
- LLM Inference : 47.47s (Load: 0.00s, Prompt Eval: 47.43s)
(TTFT: 0.041s, GPU Util: 100.0%, VRAM: 20700.0/23028.0 MiB)
(Tokens: In=1309, Out=599, speed=12.6 tok/s)
- JSON parsing : 0.00s
- Total Doc Time : 52.81s

* MA/GST Registration Certificate (GST Registration Certificate [1].pdf):
- S3 Download : 0.20s
- Azure OCR : 4.34s (Pages: 3, Avg/Page: 1.45s)
- Embedding Gen : 0.24s (Chunks: 4)
- Retrieval Time : 0.00s
- LLM Inference : 2.53s (Load: 0.00s, Prompt Eval: 2.49s)
(TTFT: 0.041s, GPU Util: 100.0%, VRAM: 20700.0/23028.0 MiB)
(Tokens: In=839, Out=68, speed=27.3 tok/s)
- JSON parsing : 0.00s
- Total Doc Time : 7.35s

----------------------------------------------------------------------
BOTTLENECK ANALYSIS & MIGRATION RECOMMENDATIONS:
- Slowest Document : CA1/UDYAM certificate (55.74s)
- Primary Bottleneck : LLM Inference (108.17s, 66.1%)

[RECOMMENDATION] Optimize LLM Inference Layer:
- If using Ollama, consider replacing it with vLLM which supports Continuous Batching and PagedAttention, increasing throughput by 3x-10x.
- If already using vLLM, optimize using tensor parallelism, model quantization (AWQ/GPTQ), or by adjusting VRAM constraints.
======================================================================
======================================================================
CREDIT-RISK LATENCY ANALYSIS & BENCHMARK REPORT: 7cfca04c-e838-4100-9bd5-ef0ed51888c6
======================================================================
Process PID : 319094
Session Start : 2026-06-02T12:15:33.210421
End-to-End Latency : 217.77 seconds
Request Status : SUCCESS
Total Documents Processed : 15
----------------------------------------------------------------------
PIPELINE STAGE LATENCY SUMMARY:
- S3 / Local Download : 2.47s ( 1.1% of total time)
- Azure OCR Engine : 39.25s ( 18.0% of total time)
- Embedding Generation : 6.01s ( 2.8% of total time)
- Vector Retrieval : 3.98s ( 1.8% of total time)
- LLM Inference : 63.76s ( 29.3% of total time)

OTHER Global Orchestration Stages:
- Verification Pipeline Time Seconds: 0.03s
- Cibil Processing Time Seconds: 0.03s
- Final Pdf Generation Time Seconds: 0.20s
- S3 Upload Time Seconds : 0.09s
- Lsq Submission Time Seconds : 3.95s
----------------------------------------------------------------------
DETAILED PER-DOCUMENT TIMING SUMMARY:
* CA1/CIBIL (CIBIL.json):
- S3 Download : 0.12s
- Azure OCR : 0.00s (Pages: 0, Avg/Page: 0.00s)
- Embedding Gen : 0.00s (Chunks: 0)
- Retrieval Time : 0.00s
- LLM Inference : 0.00s (Load: 0.00s, Prompt Eval: 0.00s)
(Tokens: In=0, Out=0, speed=0.0 tok/s)
- JSON parsing : 0.00s
- Total Doc Time : 0.17s

* Details/Details (Activity.json):
- S3 Download : 0.08s
- Azure OCR : 0.00s (Pages: 0, Avg/Page: 0.00s)
- Embedding Gen : 0.00s (Chunks: 0)
- Retrieval Time : 0.00s
- LLM Inference : 0.00s (Load: 0.00s, Prompt Eval: 0.00s)
(Tokens: In=0, Out=0, speed=0.0 tok/s)
- JSON parsing : 0.00s
- Total Doc Time : 0.14s

* Details/Details (Lead.json):
- S3 Download : 0.05s
- Azure OCR : 0.00s (Pages: 0, Avg/Page: 0.00s)
- Embedding Gen : 0.00s (Chunks: 0)
- Retrieval Time : 0.00s
- LLM Inference : 0.00s (Load: 0.00s, Prompt Eval: 0.00s)
(Tokens: In=0, Out=0, speed=0.0 tok/s)
- JSON parsing : 0.00s
- Total Doc Time : 0.05s

* KP/CIBIL (CIBIL.json):
- S3 Download : 0.10s
- Azure OCR : 0.00s (Pages: 0, Avg/Page: 0.00s)
- Embedding Gen : 0.00s (Chunks: 0)
- Retrieval Time : 0.00s
- LLM Inference : 0.00s (Load: 0.00s, Prompt Eval: 0.00s)
(Tokens: In=0, Out=0, speed=0.0 tok/s)
- JSON parsing : 0.00s
- Total Doc Time : 0.14s

* MA/CIBIL (CIBIL.json):
- S3 Download : 0.09s
- Azure OCR : 0.00s (Pages: 0, Avg/Page: 0.00s)
- Embedding Gen : 0.00s (Chunks: 0)
- Retrieval Time : 0.00s
- LLM Inference : 0.00s (Load: 0.00s, Prompt Eval: 0.00s)
(Tokens: In=0, Out=0, speed=0.0 tok/s)
- JSON parsing : 0.00s
- Total Doc Time : 0.14s

* CA1/Address (Business Address Proof [1].pdf):
- S3 Download : 0.17s
- Azure OCR : 3.39s (Pages: 3, Avg/Page: 1.13s)
- Embedding Gen : 2.85s (Chunks: 4)
- Retrieval Time : 0.00s
- LLM Inference : 1.17s (Load: 0.00s, Prompt Eval: 1.13s)
(TTFT: 0.042s, GPU Util: 100.0%, VRAM: 20211.0/23028.0 MiB)
(Tokens: In=694, Out=50, speed=44.2 tok/s)
- JSON parsing : 0.00s
- Total Doc Time : 7.62s

* CA1/UDYAM certificate (UDYAM certificate [1].pdf):
- S3 Download : 0.25s
- Azure OCR : 4.44s (Pages: 5, Avg/Page: 0.89s)
- Embedding Gen : 0.77s (Chunks: 14)
- Retrieval Time : 3.78s
- LLM Inference : 47.48s (Load: 0.00s, Prompt Eval: 47.43s)
(TTFT: 0.042s, GPU Util: 100.0%, VRAM: 20700.0/23028.0 MiB)
(Tokens: In=1592, Out=2000, speed=42.2 tok/s)
- JSON parsing : 0.00s
- Total Doc Time : 104.26s

* CA1/GST Registration Certificate (GST Registration Certificate [1].pdf):
- S3 Download : 0.22s
- Azure OCR : 3.37s (Pages: 3, Avg/Page: 1.12s)
- Embedding Gen : 0.23s (Chunks: 4)
- Retrieval Time : 0.00s
- LLM Inference : 2.53s (Load: 0.00s, Prompt Eval: 2.49s)
(TTFT: 0.043s, GPU Util: 100.0%, VRAM: 20700.0/23028.0 MiB)
(Tokens: In=1020, Out=108, speed=43.4 tok/s)
- JSON parsing : 0.00s
- Total Doc Time : 6.39s

* KP/Aadhar Card (Aadhar Card [1].pdf):
- S3 Download : 0.17s
- Azure OCR : 3.25s (Pages: 2, Avg/Page: 1.62s)
- Embedding Gen : 0.38s (Chunks: 1)
- Retrieval Time : 0.00s
- LLM Inference : 1.97s (Load: 0.00s, Prompt Eval: 1.93s)
(TTFT: 0.042s, GPU Util: 100.0%, VRAM: 20700.0/23028.0 MiB)
(Tokens: In=1203, Out=84, speed=43.5 tok/s)
- JSON parsing : 0.00s
- Total Doc Time : 5.81s

* KP/Address (Key Person POA - Aadhar [1].pdf):
- S3 Download : 0.17s
- Azure OCR : 5.33s (Pages: 2, Avg/Page: 2.67s)
- Embedding Gen : 0.19s (Chunks: 1)
- Retrieval Time : 0.00s
- LLM Inference : 1.09s (Load: 0.00s, Prompt Eval: 1.06s)
(TTFT: 0.036s, GPU Util: 100.0%, VRAM: 20700.0/23028.0 MiB)
(Tokens: In=396, Out=47, speed=44.4 tok/s)
- JSON parsing : 0.00s
- Total Doc Time : 6.82s

* KP/Address (Key Person POA - Aadhar [2].pdf):
- S3 Download : 0.19s
- Azure OCR : 5.28s (Pages: 2, Avg/Page: 2.64s)
- Embedding Gen : 0.45s (Chunks: 5)
- Retrieval Time : 0.00s
- LLM Inference : 1.18s (Load: 0.00s, Prompt Eval: 1.14s)
(TTFT: 0.039s, GPU Util: 100.0%, VRAM: 20700.0/23028.0 MiB)
(Tokens: In=1107, Out=50, speed=43.9 tok/s)
- JSON parsing : 0.00s
- Total Doc Time : 7.14s

* KP/PAN Card (PAN Card [1].pdf):
- S3 Download : 0.21s
- Azure OCR : 3.23s (Pages: 2, Avg/Page: 1.62s)
- Embedding Gen : 0.24s (Chunks: 3)
- Retrieval Time : 0.00s
- LLM Inference : 1.55s (Load: 0.00s, Prompt Eval: 1.51s)
(TTFT: 0.041s, GPU Util: 100.0%, VRAM: 20700.0/23028.0 MiB)
(Tokens: In=1110, Out=66, speed=43.7 tok/s)
- JSON parsing : 0.00s
- Total Doc Time : 5.27s

* MA/Address (Business Address Proof [1].pdf):
- S3 Download : 0.22s
- Azure OCR : 3.27s (Pages: 3, Avg/Page: 1.09s)
- Embedding Gen : 0.26s (Chunks: 4)
- Retrieval Time : 0.00s
- LLM Inference : 1.17s (Load: 0.00s, Prompt Eval: 1.13s)
(TTFT: 0.039s, GPU Util: 100.0%, VRAM: 20700.0/23028.0 MiB)
(Tokens: In=694, Out=50, speed=44.2 tok/s)
- JSON parsing : 0.00s
- Total Doc Time : 4.97s

* MA/UDYAM certificate (UDYAM certificate [1].pdf):
- S3 Download : 0.18s
- Azure OCR : 4.39s (Pages: 5, Avg/Page: 0.88s)
- Embedding Gen : 0.40s (Chunks: 14)
- Retrieval Time : 0.20s
- LLM Inference : 3.09s (Load: 0.00s, Prompt Eval: 3.04s)
(TTFT: 0.042s, GPU Util: 100.0%, VRAM: 20700.0/23028.0 MiB)
(Tokens: In=1592, Out=131, speed=43.0 tok/s)
- JSON parsing : 0.00s
- Total Doc Time : 55.80s

* MA/GST Registration Certificate (GST Registration Certificate [1].pdf):
- S3 Download : 0.24s
- Azure OCR : 3.29s (Pages: 3, Avg/Page: 1.10s)
- Embedding Gen : 0.24s (Chunks: 4)
- Retrieval Time : 0.00s
- LLM Inference : 2.53s (Load: 0.00s, Prompt Eval: 2.49s)
(TTFT: 0.041s, GPU Util: 100.0%, VRAM: 20702.0/23028.0 MiB)
(Tokens: In=1020, Out=108, speed=43.4 tok/s)
- JSON parsing : 0.00s
- Total Doc Time : 6.34s

----------------------------------------------------------------------
BOTTLENECK ANALYSIS & MIGRATION RECOMMENDATIONS:
- Slowest Document : CA1/UDYAM certificate (104.26s)
- Primary Bottleneck : LLM Inference (63.76s, 29.3%)

[RECOMMENDATION] Optimize LLM Inference Layer:
- If using Ollama, consider replacing it with vLLM which supports Continuous Batching and PagedAttention, increasing throughput by 3x-10x.
- If already using vLLM, optimize using tensor parallelism, model quantization (AWQ/GPTQ), or by adjusting VRAM constraints.
======================================================================
======================================================================
CREDIT-RISK LATENCY ANALYSIS & BENCHMARK REPORT: 874e915c-5d9d-4dce-83b4-0ece2de605a3
======================================================================
Process PID : 316494
Session Start : 2026-06-02T12:04:43.387534
End-to-End Latency : 119.77 seconds
Request Status : SUCCESS
Total Documents Processed : 15
----------------------------------------------------------------------
PIPELINE STAGE LATENCY SUMMARY:
- S3 / Local Download : 2.55s ( 2.1% of total time)
- Azure OCR Engine : 39.58s ( 33.0% of total time)
- Embedding Generation : 2.65s ( 2.2% of total time)
- Vector Retrieval : 4.18s ( 3.5% of total time)
- LLM Inference : 19.23s ( 16.1% of total time)

OTHER Global Orchestration Stages:
- Verification Pipeline Time Seconds: 0.03s
- Cibil Processing Time Seconds: 0.03s
- Final Pdf Generation Time Seconds: 0.19s
- S3 Upload Time Seconds : 0.10s
- Lsq Submission Time Seconds : 0.69s
----------------------------------------------------------------------
DETAILED PER-DOCUMENT TIMING SUMMARY:
* CA1/CIBIL (CIBIL.json):
- S3 Download : 0.13s
- Azure OCR : 0.00s (Pages: 0, Avg/Page: 0.00s)
- Embedding Gen : 0.00s (Chunks: 0)
- Retrieval Time : 0.00s
- LLM Inference : 0.00s (Load: 0.00s, Prompt Eval: 0.00s)
(Tokens: In=0, Out=0, speed=0.0 tok/s)
- JSON parsing : 0.00s
- Total Doc Time : 0.18s

* Details/Details (Activity.json):
- S3 Download : 0.10s
- Azure OCR : 0.00s (Pages: 0, Avg/Page: 0.00s)
- Embedding Gen : 0.00s (Chunks: 0)
- Retrieval Time : 0.00s
- LLM Inference : 0.00s (Load: 0.00s, Prompt Eval: 0.00s)
(Tokens: In=0, Out=0, speed=0.0 tok/s)
- JSON parsing : 0.00s
- Total Doc Time : 0.17s

* Details/Details (Lead.json):
- S3 Download : 0.06s
- Azure OCR : 0.00s (Pages: 0, Avg/Page: 0.00s)
- Embedding Gen : 0.00s (Chunks: 0)
- Retrieval Time : 0.00s
- LLM Inference : 0.00s (Load: 0.00s, Prompt Eval: 0.00s)
(Tokens: In=0, Out=0, speed=0.0 tok/s)
- JSON parsing : 0.00s
- Total Doc Time : 0.06s

* KP/CIBIL (CIBIL.json):
- S3 Download : 0.11s
- Azure OCR : 0.00s (Pages: 0, Avg/Page: 0.00s)
- Embedding Gen : 0.00s (Chunks: 0)
- Retrieval Time : 0.00s
- LLM Inference : 0.00s (Load: 0.00s, Prompt Eval: 0.00s)
(Tokens: In=0, Out=0, speed=0.0 tok/s)
- JSON parsing : 0.00s
- Total Doc Time : 0.16s

* MA/CIBIL (CIBIL.json):
- S3 Download : 0.09s
- Azure OCR : 0.00s (Pages: 0, Avg/Page: 0.00s)
- Embedding Gen : 0.00s (Chunks: 0)
- Retrieval Time : 0.00s
- LLM Inference : 0.00s (Load: 0.00s, Prompt Eval: 0.00s)
(Tokens: In=0, Out=0, speed=0.0 tok/s)
- JSON parsing : 0.00s
- Total Doc Time : 0.13s

* CA1/Address (Business Address Proof [1].pdf):
- S3 Download : 0.18s
- Azure OCR : 3.42s (Pages: 3, Avg/Page: 1.14s)
- Embedding Gen : 0.28s (Chunks: 4)
- Retrieval Time : 0.00s
- LLM Inference : 1.18s (Load: 0.00s, Prompt Eval: 1.13s)
(TTFT: 0.043s, GPU Util: 100.0%, VRAM: 20213.0/23028.0 MiB)
(Tokens: In=694, Out=50, speed=44.2 tok/s)
- JSON parsing : 0.00s
- Total Doc Time : 5.09s

* CA1/UDYAM certificate (UDYAM certificate [1].pdf):
- S3 Download : 0.24s
- Azure OCR : 4.47s (Pages: 5, Avg/Page: 0.89s)
- Embedding Gen : 0.40s (Chunks: 14)
- Retrieval Time : 3.97s
- LLM Inference : 2.92s (Load: 0.00s, Prompt Eval: 2.88s)
(TTFT: 0.042s, GPU Util: 100.0%, VRAM: 20702.0/23028.0 MiB)
(Tokens: In=1592, Out=124, speed=43.1 tok/s)
- JSON parsing : 0.00s
- Total Doc Time : 59.54s

* CA1/GST Registration Certificate (GST Registration Certificate [1].pdf):
- S3 Download : 0.25s
- Azure OCR : 3.33s (Pages: 3, Avg/Page: 1.11s)
- Embedding Gen : 0.25s (Chunks: 4)
- Retrieval Time : 0.00s
- LLM Inference : 2.53s (Load: 0.00s, Prompt Eval: 2.49s)
(TTFT: 0.041s, GPU Util: 100.0%, VRAM: 20702.0/23028.0 MiB)
(Tokens: In=1020, Out=108, speed=43.4 tok/s)
- JSON parsing : 0.00s
- Total Doc Time : 6.40s

* KP/Aadhar Card (Aadhar Card [1].pdf):
- S3 Download : 0.18s
- Azure OCR : 3.33s (Pages: 2, Avg/Page: 1.66s)
- Embedding Gen : 0.19s (Chunks: 1)
- Retrieval Time : 0.00s
- LLM Inference : 1.97s (Load: 0.00s, Prompt Eval: 1.93s)
(TTFT: 0.041s, GPU Util: 100.0%, VRAM: 20702.0/23028.0 MiB)
(Tokens: In=1203, Out=84, speed=43.5 tok/s)
- JSON parsing : 0.00s
- Total Doc Time : 5.70s

* KP/Address (Key Person POA - Aadhar [1].pdf):
- S3 Download : 0.21s
- Azure OCR : 3.24s (Pages: 2, Avg/Page: 1.62s)
- Embedding Gen : 0.19s (Chunks: 1)
- Retrieval Time : 0.00s
- LLM Inference : 1.12s (Load: 0.00s, Prompt Eval: 1.08s)
(TTFT: 0.036s, GPU Util: 100.0%, VRAM: 20702.0/23028.0 MiB)
(Tokens: In=396, Out=48, speed=44.4 tok/s)
- JSON parsing : 0.00s
- Total Doc Time : 4.80s

* KP/Address (Key Person POA - Aadhar [2].pdf):
- S3 Download : 0.15s
- Azure OCR : 4.28s (Pages: 2, Avg/Page: 2.14s)
- Embedding Gen : 0.25s (Chunks: 5)
- Retrieval Time : 0.00s
- LLM Inference : 1.18s (Load: 0.00s, Prompt Eval: 1.14s)
(TTFT: 0.039s, GPU Util: 100.0%, VRAM: 20702.0/23028.0 MiB)
(Tokens: In=1107, Out=50, speed=43.9 tok/s)
- JSON parsing : 0.00s
- Total Doc Time : 5.89s

* KP/PAN Card (PAN Card [1].pdf):
- S3 Download : 0.20s
- Azure OCR : 4.41s (Pages: 2, Avg/Page: 2.21s)
- Embedding Gen : 0.22s (Chunks: 3)
- Retrieval Time : 0.00s
- LLM Inference : 1.55s (Load: 0.00s, Prompt Eval: 1.51s)
(TTFT: 0.041s, GPU Util: 100.0%, VRAM: 20702.0/23028.0 MiB)
(Tokens: In=1110, Out=66, speed=43.7 tok/s)
- JSON parsing : 0.00s
- Total Doc Time : 6.43s

* MA/Address (Business Address Proof [1].pdf):
- S3 Download : 0.21s
- Azure OCR : 4.31s (Pages: 3, Avg/Page: 1.44s)
- Embedding Gen : 0.24s (Chunks: 4)
- Retrieval Time : 0.00s
- LLM Inference : 1.17s (Load: 0.00s, Prompt Eval: 1.13s)
(TTFT: 0.038s, GPU Util: 97.0%, VRAM: 20702.0/23028.0 MiB)
(Tokens: In=694, Out=50, speed=44.2 tok/s)
- JSON parsing : 0.00s
- Total Doc Time : 5.97s

* MA/UDYAM certificate (UDYAM certificate [1].pdf):
- S3 Download : 0.21s
- Azure OCR : 3.39s (Pages: 5, Avg/Page: 0.68s)
- Embedding Gen : 0.39s (Chunks: 14)
- Retrieval Time : 0.21s
- LLM Inference : 3.09s (Load: 0.00s, Prompt Eval: 3.04s)
(TTFT: 0.042s, GPU Util: 100.0%, VRAM: 20702.0/23028.0 MiB)
(Tokens: In=1592, Out=131, speed=43.0 tok/s)
- JSON parsing : 0.00s
- Total Doc Time : 7.32s

* MA/GST Registration Certificate (GST Registration Certificate [1].pdf):
- S3 Download : 0.21s
- Azure OCR : 5.42s (Pages: 3, Avg/Page: 1.81s)
- Embedding Gen : 0.25s (Chunks: 4)
- Retrieval Time : 0.00s
- LLM Inference : 2.53s (Load: 0.00s, Prompt Eval: 2.49s)
(TTFT: 0.042s, GPU Util: 100.0%, VRAM: 20702.0/23028.0 MiB)
(Tokens: In=1020, Out=108, speed=43.4 tok/s)
- JSON parsing : 0.00s
- Total Doc Time : 8.44s

----------------------------------------------------------------------
BOTTLENECK ANALYSIS & MIGRATION RECOMMENDATIONS:
- Slowest Document : CA1/UDYAM certificate (59.54s)
- Primary Bottleneck : Azure OCR Engine (39.58s, 33.0%)

[RECOMMENDATION] Optimize Azure Document Intelligence:
- Use the 'prebuilt-read' model instead of 'prebuilt-layout' for simple structured documents (Aadhaar, PAN) as it is twice as fast.
- If document lengths are high, run asynchronous table/layout extractions in parallel worker pools to hide polling delays.
======================================================================
======================================================================
CREDIT-RISK LATENCY ANALYSIS & BENCHMARK REPORT: e6f026c3-92af-4911-b7ca-0442866f68e7
======================================================================
Process PID : 313486
Session Start : 2026-06-02T12:01:37.002952
End-to-End Latency : 115.05 seconds
Request Status : SUCCESS
Total Documents Processed : 15
----------------------------------------------------------------------
PIPELINE STAGE LATENCY SUMMARY:
- S3 / Local Download : 2.61s ( 2.3% of total time)
- Azure OCR Engine : 35.37s ( 30.7% of total time)
- Embedding Generation : 5.51s ( 4.8% of total time)
- Vector Retrieval : 4.16s ( 3.6% of total time)
- LLM Inference : 63.60s ( 55.3% of total time)

OTHER Global Orchestration Stages:
- Verification Pipeline Time Seconds: 0.03s
- Cibil Processing Time Seconds: 0.03s
- Final Pdf Generation Time Seconds: 0.20s
- S3 Upload Time Seconds : 0.10s
- Lsq Submission Time Seconds : 0.61s
----------------------------------------------------------------------
DETAILED PER-DOCUMENT TIMING SUMMARY:
* CA1/CIBIL (CIBIL.json):
- S3 Download : 0.14s
- Azure OCR : 0.00s (Pages: 0, Avg/Page: 0.00s)
- Embedding Gen : 0.00s (Chunks: 0)
- Retrieval Time : 0.00s
- LLM Inference : 0.00s (Load: 0.00s, Prompt Eval: 0.00s)
(Tokens: In=0, Out=0, speed=0.0 tok/s)
- JSON parsing : 0.00s
- Total Doc Time : 0.20s

* Details/Details (Activity.json):
- S3 Download : 0.09s
- Azure OCR : 0.00s (Pages: 0, Avg/Page: 0.00s)
- Embedding Gen : 0.00s (Chunks: 0)
- Retrieval Time : 0.00s
- LLM Inference : 0.00s (Load: 0.00s, Prompt Eval: 0.00s)
(Tokens: In=0, Out=0, speed=0.0 tok/s)
- JSON parsing : 0.00s
- Total Doc Time : 0.15s

* Details/Details (Lead.json):
- S3 Download : 0.03s
- Azure OCR : 0.00s (Pages: 0, Avg/Page: 0.00s)
- Embedding Gen : 0.00s (Chunks: 0)
- Retrieval Time : 0.00s
- LLM Inference : 0.00s (Load: 0.00s, Prompt Eval: 0.00s)
(Tokens: In=0, Out=0, speed=0.0 tok/s)
- JSON parsing : 0.00s
- Total Doc Time : 0.03s

* KP/CIBIL (CIBIL.json):
- S3 Download : 0.10s
- Azure OCR : 0.00s (Pages: 0, Avg/Page: 0.00s)
- Embedding Gen : 0.00s (Chunks: 0)
- Retrieval Time : 0.00s
- LLM Inference : 0.00s (Load: 0.00s, Prompt Eval: 0.00s)
(Tokens: In=0, Out=0, speed=0.0 tok/s)
- JSON parsing : 0.00s
- Total Doc Time : 0.14s

* MA/CIBIL (CIBIL.json):
- S3 Download : 0.09s
- Azure OCR : 0.00s (Pages: 0, Avg/Page: 0.00s)
- Embedding Gen : 0.00s (Chunks: 0)
- Retrieval Time : 0.00s
- LLM Inference : 0.00s (Load: 0.00s, Prompt Eval: 0.00s)
(Tokens: In=0, Out=0, speed=0.0 tok/s)
- JSON parsing : 0.00s
- Total Doc Time : 0.13s

* CA1/Address (Business Address Proof [1].pdf):
- S3 Download : 0.22s
- Azure OCR : 3.47s (Pages: 3, Avg/Page: 1.16s)
- Embedding Gen : 2.59s (Chunks: 4)
- Retrieval Time : 0.00s
- LLM Inference : 1.18s (Load: 0.00s, Prompt Eval: 1.13s)
(TTFT: 0.047s, GPU Util: 100.0%, VRAM: 20700.0/23028.0 MiB)
(Tokens: In=694, Out=50, speed=44.2 tok/s)
- JSON parsing : 0.00s
- Total Doc Time : 7.50s

* CA1/UDYAM certificate (UDYAM certificate [1].pdf):
- S3 Download : 0.23s
- Azure OCR : 4.42s (Pages: 5, Avg/Page: 0.88s)
- Embedding Gen : 0.57s (Chunks: 14)
- Retrieval Time : 3.95s
- LLM Inference : 3.11s (Load: 0.00s, Prompt Eval: 3.07s)
(TTFT: 0.042s, GPU Util: 100.0%, VRAM: 21190.0/23028.0 MiB)
(Tokens: In=1592, Out=132, speed=43.0 tok/s)
- JSON parsing : 0.00s
- Total Doc Time : 12.32s

* CA1/GST Registration Certificate (GST Registration Certificate [1].pdf):
- S3 Download : 0.23s
- Azure OCR : 3.38s (Pages: 3, Avg/Page: 1.13s)
- Embedding Gen : 0.23s (Chunks: 4)
- Retrieval Time : 0.00s
- LLM Inference : 2.53s (Load: 0.00s, Prompt Eval: 2.49s)
(TTFT: 0.041s, GPU Util: 100.0%, VRAM: 21190.0/23028.0 MiB)
(Tokens: In=1020, Out=108, speed=43.4 tok/s)
- JSON parsing : 0.00s
- Total Doc Time : 6.40s

* KP/Aadhar Card (Aadhar Card [1].pdf):
- S3 Download : 0.22s
- Azure OCR : 3.31s (Pages: 2, Avg/Page: 1.66s)
- Embedding Gen : 0.37s (Chunks: 1)
- Retrieval Time : 0.00s
- LLM Inference : 1.97s (Load: 0.00s, Prompt Eval: 1.93s)
(TTFT: 0.042s, GPU Util: 100.0%, VRAM: 21190.0/23028.0 MiB)
(Tokens: In=1203, Out=84, speed=43.5 tok/s)
- JSON parsing : 0.00s
- Total Doc Time : 5.92s

* KP/Address (Key Person POA - Aadhar [1].pdf):
- S3 Download : 0.23s
- Azure OCR : 3.22s (Pages: 2, Avg/Page: 1.61s)
- Embedding Gen : 0.19s (Chunks: 1)
- Retrieval Time : 0.00s
- LLM Inference : 0.91s (Load: 0.00s, Prompt Eval: 0.87s)
(TTFT: 0.036s, GPU Util: 100.0%, VRAM: 21190.0/23028.0 MiB)
(Tokens: In=396, Out=39, speed=44.6 tok/s)
- JSON parsing : 0.00s
- Total Doc Time : 4.59s

* KP/Address (Key Person POA - Aadhar [2].pdf):
- S3 Download : 0.16s
- Azure OCR : 4.28s (Pages: 2, Avg/Page: 2.14s)
- Embedding Gen : 0.47s (Chunks: 5)
- Retrieval Time : 0.00s
- LLM Inference : 1.18s (Load: 0.00s, Prompt Eval: 1.14s)
(TTFT: 0.039s, GPU Util: 100.0%, VRAM: 21190.0/23028.0 MiB)
(Tokens: In=1107, Out=50, speed=43.9 tok/s)
- JSON parsing : 0.00s
- Total Doc Time : 6.11s

* KP/PAN Card (PAN Card [1].pdf):
- S3 Download : 0.18s
- Azure OCR : 3.29s (Pages: 2, Avg/Page: 1.64s)
- Embedding Gen : 0.22s (Chunks: 3)
- Retrieval Time : 0.00s
- LLM Inference : 1.55s (Load: 0.00s, Prompt Eval: 1.51s)
(TTFT: 0.041s, GPU Util: 100.0%, VRAM: 21190.0/23028.0 MiB)
(Tokens: In=1110, Out=66, speed=43.7 tok/s)
- JSON parsing : 0.00s
- Total Doc Time : 5.27s

* MA/Address (Business Address Proof [1].pdf):
- S3 Download : 0.20s
- Azure OCR : 3.26s (Pages: 3, Avg/Page: 1.09s)
- Embedding Gen : 0.23s (Chunks: 4)
- Retrieval Time : 0.00s
- LLM Inference : 1.17s (Load: 0.00s, Prompt Eval: 1.13s)
(TTFT: 0.040s, GPU Util: 100.0%, VRAM: 21190.0/23028.0 MiB)
(Tokens: In=694, Out=50, speed=44.2 tok/s)
- JSON parsing : 0.00s
- Total Doc Time : 4.90s

* MA/UDYAM certificate (UDYAM certificate [1].pdf):
- S3 Download : 0.26s
- Azure OCR : 3.41s (Pages: 5, Avg/Page: 0.68s)
- Embedding Gen : 0.41s (Chunks: 14)
- Retrieval Time : 0.21s
- LLM Inference : 47.48s (Load: 0.00s, Prompt Eval: 47.44s)
(TTFT: 0.043s, GPU Util: 100.0%, VRAM: 21192.0/23028.0 MiB)
(Tokens: In=1592, Out=2000, speed=42.2 tok/s)
- JSON parsing : 0.00s
- Total Doc Time : 51.81s

* MA/GST Registration Certificate (GST Registration Certificate [1].pdf):
- S3 Download : 0.24s
- Azure OCR : 3.34s (Pages: 3, Avg/Page: 1.11s)
- Embedding Gen : 0.25s (Chunks: 4)
- Retrieval Time : 0.00s
- LLM Inference : 2.53s (Load: 0.00s, Prompt Eval: 2.49s)
(TTFT: 0.041s, GPU Util: 98.0%, VRAM: 21192.0/23028.0 MiB)
(Tokens: In=1020, Out=108, speed=43.4 tok/s)
- JSON parsing : 0.00s
- Total Doc Time : 6.39s

----------------------------------------------------------------------
BOTTLENECK ANALYSIS & MIGRATION RECOMMENDATIONS:
- Slowest Document : MA/UDYAM certificate (51.81s)
- Primary Bottleneck : LLM Inference (63.60s, 55.3%)

[RECOMMENDATION] Optimize LLM Inference Layer:
- If using Ollama, consider replacing it with vLLM which supports Continuous Batching and PagedAttention, increasing throughput by 3x-10x.
- If already using vLLM, optimize using tensor parallelism, model quantization (AWQ/GPTQ), or by adjusting VRAM constraints.
======================================================================
======================================================================
CREDIT-RISK LATENCY ANALYSIS & BENCHMARK REPORT: 66f9f227-40a6-4d86-a7fc-4e5d031be6cd
======================================================================
Process PID : 313476
Session Start : 2026-06-02T11:50:21.263287
End-to-End Latency : 253.47 seconds
Request Status : SUCCESS
Total Documents Processed : 15
----------------------------------------------------------------------
PIPELINE STAGE LATENCY SUMMARY:
- S3 / Local Download : 2.38s ( 0.9% of total time)
- Azure OCR Engine : 34.31s ( 13.5% of total time)
- Embedding Generation : 2.70s ( 1.1% of total time)
- Vector Retrieval : 4.06s ( 1.6% of total time)
- LLM Inference : 107.96s ( 42.6% of total time)

OTHER Global Orchestration Stages:
- Verification Pipeline Time Seconds: 0.03s
- Cibil Processing Time Seconds: 0.03s
- Final Pdf Generation Time Seconds: 0.19s
- S3 Upload Time Seconds : 0.10s
- Lsq Submission Time Seconds : 3.80s
----------------------------------------------------------------------
DETAILED PER-DOCUMENT TIMING SUMMARY:
* CA1/CIBIL (CIBIL.json):
- S3 Download : 0.12s
- Azure OCR : 0.00s (Pages: 0, Avg/Page: 0.00s)
- Embedding Gen : 0.00s (Chunks: 0)
- Retrieval Time : 0.00s
- LLM Inference : 0.00s (Load: 0.00s, Prompt Eval: 0.00s)
(Tokens: In=0, Out=0, speed=0.0 tok/s)
- JSON parsing : 0.00s
- Total Doc Time : 0.17s

* Details/Details (Activity.json):
- S3 Download : 0.11s
- Azure OCR : 0.00s (Pages: 0, Avg/Page: 0.00s)
- Embedding Gen : 0.00s (Chunks: 0)
- Retrieval Time : 0.00s
- LLM Inference : 0.00s (Load: 0.00s, Prompt Eval: 0.00s)
(Tokens: In=0, Out=0, speed=0.0 tok/s)
- JSON parsing : 0.00s
- Total Doc Time : 0.17s

* Details/Details (Lead.json):
- S3 Download : 0.03s
- Azure OCR : 0.00s (Pages: 0, Avg/Page: 0.00s)
- Embedding Gen : 0.00s (Chunks: 0)
- Retrieval Time : 0.00s
- LLM Inference : 0.00s (Load: 0.00s, Prompt Eval: 0.00s)
(Tokens: In=0, Out=0, speed=0.0 tok/s)
- JSON parsing : 0.00s
- Total Doc Time : 0.03s

* KP/CIBIL (CIBIL.json):
- S3 Download : 0.09s
- Azure OCR : 0.00s (Pages: 0, Avg/Page: 0.00s)
- Embedding Gen : 0.00s (Chunks: 0)
- Retrieval Time : 0.00s
- LLM Inference : 0.00s (Load: 0.00s, Prompt Eval: 0.00s)
(Tokens: In=0, Out=0, speed=0.0 tok/s)
- JSON parsing : 0.00s
- Total Doc Time : 0.13s

* MA/CIBIL (CIBIL.json):
- S3 Download : 0.10s
- Azure OCR : 0.00s (Pages: 0, Avg/Page: 0.00s)
- Embedding Gen : 0.00s (Chunks: 0)
- Retrieval Time : 0.00s
- LLM Inference : 0.00s (Load: 0.00s, Prompt Eval: 0.00s)
(Tokens: In=0, Out=0, speed=0.0 tok/s)
- JSON parsing : 0.00s
- Total Doc Time : 0.15s

* CA1/Address (Business Address Proof [1].pdf):
- S3 Download : 0.19s
- Azure OCR : 3.35s (Pages: 3, Avg/Page: 1.12s)
- Embedding Gen : 0.26s (Chunks: 4)
- Retrieval Time : 0.00s
- LLM Inference : 1.17s (Load: 0.00s, Prompt Eval: 1.13s)
(TTFT: 0.043s, GPU Util: 100.0%, VRAM: 20213.0/23028.0 MiB)
(Tokens: In=694, Out=50, speed=44.2 tok/s)
- JSON parsing : 0.00s
- Total Doc Time : 5.01s

* CA1/UDYAM certificate (UDYAM certificate [1].pdf):
- S3 Download : 0.21s
- Azure OCR : 3.40s (Pages: 5, Avg/Page: 0.68s)
- Embedding Gen : 0.40s (Chunks: 14)
- Retrieval Time : 3.87s
- LLM Inference : 47.48s (Load: 0.00s, Prompt Eval: 47.44s)
(TTFT: 0.042s, GPU Util: 82.0%, VRAM: 20702.0/23028.0 MiB)
(Tokens: In=1592, Out=2000, speed=42.2 tok/s)
- JSON parsing : 0.00s
- Total Doc Time : 150.41s

* CA1/GST Registration Certificate (GST Registration Certificate [1].pdf):
- S3 Download : 0.23s
- Azure OCR : 3.34s (Pages: 3, Avg/Page: 1.11s)
- Embedding Gen : 0.29s (Chunks: 4)
- Retrieval Time : 0.00s
- LLM Inference : 2.53s (Load: 0.00s, Prompt Eval: 2.49s)
(TTFT: 0.041s, GPU Util: 100.0%, VRAM: 20702.0/23028.0 MiB)
(Tokens: In=1020, Out=108, speed=43.4 tok/s)
- JSON parsing : 0.00s
- Total Doc Time : 6.43s

* KP/Aadhar Card (Aadhar Card [1].pdf):
- S3 Download : 0.20s
- Azure OCR : 3.25s (Pages: 2, Avg/Page: 1.62s)
- Embedding Gen : 0.19s (Chunks: 1)
- Retrieval Time : 0.00s
- LLM Inference : 1.97s (Load: 0.00s, Prompt Eval: 1.93s)
(TTFT: 0.042s, GPU Util: 100.0%, VRAM: 20702.0/23028.0 MiB)
(Tokens: In=1203, Out=84, speed=43.5 tok/s)
- JSON parsing : 0.00s
- Total Doc Time : 5.65s

* KP/Address (Key Person POA - Aadhar [1].pdf):
- S3 Download : 0.19s
- Azure OCR : 3.41s (Pages: 2, Avg/Page: 1.70s)
- Embedding Gen : 0.20s (Chunks: 1)
- Retrieval Time : 0.00s
- LLM Inference : 0.91s (Load: 0.00s, Prompt Eval: 0.87s)
(TTFT: 0.036s, GPU Util: 100.0%, VRAM: 20702.0/23028.0 MiB)
(Tokens: In=396, Out=39, speed=44.6 tok/s)
- JSON parsing : 0.00s
- Total Doc Time : 4.74s

* KP/Address (Key Person POA - Aadhar [2].pdf):
- S3 Download : 0.16s
- Azure OCR : 3.25s (Pages: 2, Avg/Page: 1.62s)
- Embedding Gen : 0.28s (Chunks: 5)
- Retrieval Time : 0.00s
- LLM Inference : 1.18s (Load: 0.00s, Prompt Eval: 1.14s)
(TTFT: 0.039s, GPU Util: 100.0%, VRAM: 20702.0/23028.0 MiB)
(Tokens: In=1107, Out=50, speed=43.9 tok/s)
- JSON parsing : 0.00s
- Total Doc Time : 4.91s

* KP/PAN Card (PAN Card [1].pdf):
- S3 Download : 0.17s
- Azure OCR : 3.30s (Pages: 2, Avg/Page: 1.65s)
- Embedding Gen : 0.22s (Chunks: 3)
- Retrieval Time : 0.00s
- LLM Inference : 1.55s (Load: 0.00s, Prompt Eval: 1.51s)
(TTFT: 0.041s, GPU Util: 100.0%, VRAM: 20702.0/23028.0 MiB)
(Tokens: In=1110, Out=66, speed=43.7 tok/s)
- JSON parsing : 0.00s
- Total Doc Time : 5.27s

* MA/Address (Business Address Proof [1].pdf):
- S3 Download : 0.19s
- Azure OCR : 3.26s (Pages: 3, Avg/Page: 1.09s)
- Embedding Gen : 0.23s (Chunks: 4)
- Retrieval Time : 0.00s
- LLM Inference : 1.17s (Load: 0.00s, Prompt Eval: 1.13s)
(TTFT: 0.039s, GPU Util: 100.0%, VRAM: 20702.0/23028.0 MiB)
(Tokens: In=694, Out=50, speed=44.2 tok/s)
- JSON parsing : 0.00s
- Total Doc Time : 4.89s

* MA/UDYAM certificate (UDYAM certificate [1].pdf):
- S3 Download : 0.18s
- Azure OCR : 4.50s (Pages: 5, Avg/Page: 0.90s)
- Embedding Gen : 0.42s (Chunks: 14)
- Retrieval Time : 0.20s
- LLM Inference : 47.47s (Load: 0.00s, Prompt Eval: 47.43s)
(TTFT: 0.042s, GPU Util: 100.0%, VRAM: 20702.0/23028.0 MiB)
(Tokens: In=1592, Out=2000, speed=42.2 tok/s)
- JSON parsing : 0.00s
- Total Doc Time : 52.80s

* MA/GST Registration Certificate (GST Registration Certificate [1].pdf):
- S3 Download : 0.20s
- Azure OCR : 3.26s (Pages: 3, Avg/Page: 1.09s)
- Embedding Gen : 0.22s (Chunks: 4)
- Retrieval Time : 0.00s
- LLM Inference : 2.53s (Load: 0.00s, Prompt Eval: 2.49s)
(TTFT: 0.041s, GPU Util: 100.0%, VRAM: 20702.0/23028.0 MiB)
(Tokens: In=1020, Out=108, speed=43.4 tok/s)
- JSON parsing : 0.00s
- Total Doc Time : 6.25s

----------------------------------------------------------------------
BOTTLENECK ANALYSIS & MIGRATION RECOMMENDATIONS:
- Slowest Document : CA1/UDYAM certificate (150.41s)
- Primary Bottleneck : LLM Inference (107.96s, 42.6%)

[RECOMMENDATION] Optimize LLM Inference Layer:
- If using Ollama, consider replacing it with vLLM which supports Continuous Batching and PagedAttention, increasing throughput by 3x-10x.
- If already using vLLM, optimize using tensor parallelism, model quantization (AWQ/GPTQ), or by adjusting VRAM constraints.
======================================================================