ev-battery-agent

An autonomous AI agent built in Go that monitors Rivian EV battery health, detects safety threshold violations using Gemini 2.0 Flash + RAG over owner manuals, and auto-files Jira tickets directly within your Atlassian Jira workspace

If a problem occurs (such as a battery thermal risk or a voltage anomaly) the agent creates a high-priority Atlassian Jira ticket so engineers can act before it leads to a recall.

How it works

1. Monitor — Submit a battery report in plain English (TUI) or structured telemetry JSON/CSV (REST API). Readings include temperature, voltage, state of charge, and driving mode.
2. Think — The agent retrieves relevant chunks from Rivian R1S/R1T owner manuals via RAG (cosine similarity over Vertex AI embeddings), then calls Gemini 2.0 Flash to reason about whether any safety thresholds are violated.
3. Act — If a violation is detected, the agent calls the fileEngineeringTicket tool, which creates a Jira issue with severity-based priority (CRITICAL → Bug/Highest, WARNING → Task/High, INFO → Task/Medium). Duplicate tickets for the same VIN are skipped automatically.

Stack

Layer	Technology
Language	Go 1.22
LLM	Gemini 2.0 Flash (Vertex AI)
Embeddings	text-embedding-004 (Vertex AI)
LLM framework	tmc/langchaingo
Jira client	andygrunwald/go-jira
TUI	charmbracelet/bubbletea + lipgloss
RAG	In-memory cosine similarity vector store

Install

go install github.com/OpenAgents-Illinois/ev-battery-agent@latest

Or clone and build manually:

git clone https://github.com/OpenAgents-Illinois/ev-battery-agent
cd ev-battery-agent
go build -o ev-battery-agent .

Setup

2. Create a .env file

GCLOUD_PROJECT_ID=your-gcp-project
JIRA_DOMAIN=yourname.atlassian.net
JIRA_EMAIL=you@example.com
JIRA_TOKEN=your-atlassian-api-token
JIRA_SPACE_KEY=KAN

3. Authenticate with GCP

gcloud auth application-default login

Build & Run

go build -o ev-battery-agent .  # if cloned manually

# Bubble Tea TUI (interactive mode)
./ev-battery-agent

# HTTP server on port 8080
./ev-battery-agent --server

# HTTP server on custom port
./ev-battery-agent --server 9090

First run embeds the Rivian owner manual PDFs via Vertex AI and caches them locally (~30-60s). Every subsequent run loads from cache and starts in seconds.

REST API

# Analyze telemetry (JSON)
curl -X POST http://localhost:8080/analyze \
  -H "Content-Type: application/json" \
  -d '{"vin":"VIN_789","batteryTempC":55.0,"voltageV":3.1,"stateOfChargePercent":82.0,"drivingMode":"driving"}'

# Analyze telemetry (CSV: vin,tempC,voltageV[,soc%][,mode])
curl -X POST http://localhost:8080/analyze \
  -H "Content-Type: text/plain" \
  -d 'VIN_789,55.0,3.1,82.0,driving'

# Health check
curl http://localhost:8080/health