Evidence-grounded agentic RAG · reliability intelligence
Root-cause investigation with evidence, tools, verification, and human control.
ReliabilityOps combines hybrid retrieval, structured analytics, a bounded LangGraph investigation workflow, local Qwen3 reasoning, evidence verification, and human approval for industrial reliability investigations.
Hosted portfolio demo: this Static Space replays a genuine precomputed investigation from the validated local V6 pipeline. Static hosting does not execute Qwen3, BGE-M3, LangGraph, Qdrant, PostgreSQL, or Python inference live. The full reproducible implementation is available on GitHub.
Investigation demo
Inspect a validated V6 investigation
Investigation Request
Ready to investigate
Run the demonstration to replay the validated V6 workflow.
Retrieving historical incidents...
Replaying validated investigation stages
Primary hypothesis
Investigation Timeline
Evidence
Run the demonstration to reveal cited evidence.
Analytical Findings
Structured comparisons appear after the demo runs.
How it works
One bounded investigation system
Hybrid Retrieval
BGE-M3 + BM25 + RRF
BGE-M3 + BM25 + RRF
LangGraph Agent
bounded retries
bounded retries
Analytical Tools
SQL + statistics
SQL + statistics
Qwen3
local synthesis
local synthesis
Verifier
claims + citations
claims + citations
Human Review
approve / reject
approve / reject
Measured evolution
Evaluation drove the architecture
Retrieval Hit@10 · development
RCA Top-1 · development