2024 - 2025
LLM Telemetry Copilot for Fleet Analytics
Created an AI assistant that translated natural-language questions into structured telemetry insights, improving analyst speed and engineering leverage.
LangChainLangGraphRAGvLLMPythonTypeScript
Engineer Adoption
8 engineers
Daily Query Volume
600+
Infra Cost Reduction
$1,600/month
Goal
Enable non-ML stakeholders to query trip and fleet behavior without writing SQL or navigating dashboards.
Implementation
I built a retrieval-augmented agent pipeline with role-specific prompts and context-aware query planning.
- Modeled task routing with LangGraph state transitions for question classification and execution strategy.
- Added retrieval over telemetry summaries and key metrics for grounded responses.
- Deployed on on-prem GPU inference endpoints to control cost and latency.
Impact
The assistant became a practical internal tool for exploratory analysis and rapid triage.
Future Extensions
- Add feedback loops for answer-quality scoring.
- Integrate chart generation for trend-heavy questions.
- Support deeper comparative analytics between driver cohorts.