Corpus & Knowledge Scale
Verified
Indexed Vectors
1,000,233
10.002 Lakh Chunks
Parent Documents
138,079
Full context articles
Languages
HI (46k) · KN (46k) · EN (46k)
Multilingual Parquet
Embedding Model
MiniLM-L12-v2 (CUDA)
384-Dim Dense · Cosine IP
Latency Benchmark Percentiles
Sub-200ms Target
| Step |
Pipeline Stage |
P50 |
P70 |
P100 |
| 01 |
Transport & Safety Guardrail |
0.5 ms |
0.8 ms |
1.2 ms |
| 02 |
CUDA Dense Embedding |
32.0 ms |
35.0 ms |
45.0 ms |
| 03 |
FAISS 1M Vector Search |
150.0 ms |
165.0 ms |
185.0 ms |
| 04 |
Small-to-Big Parent Resolution |
2.0 ms |
2.5 ms |
5.0 ms |
| 05 |
Groq LPU LLM Generation |
190.0 ms |
350.0 ms |
600.0 ms |
Guardrail Qualification Suite
100% Passed
🛡️ Abuse & Profanity Filter
QUALIFIED
🎯 Fast-Reject Off-Topic Gate
QUALIFIED
🔗 Grounding & Anti-Hallucination
QUALIFIED
🌐 Multilingual Language Alignment
QUALIFIED
Cryptographic Artifact Hashes
SHA-256
vector_index.faiss 1,465 MB
1d1189b240c76143fd6dcdbfea02a8eb5c83051a5bdb368e0969266b90371071
parent_documents.json 526 MB
65d2deb4796ef022b56f2f32cd2d9009a5f7d8c94e0ea9896ccc6a04dbb00f08
hintrain.parquet 3,547 MB
dd88b273ad1ba381ae44bb0fc318e48bde20357b5fabce5211d62b343ec1e700
kantrain.parquet 3,712 MB
8c554042ae9c4c4409bc3b177515eca7729975de4726e319fa2ac82597e35726
ingestion_state.json 14.1 KB
cbe111d5656183600fc6cdaeb49b0b3cf18af4123c68abeb762c5edff6fb73f9