geniable 2.2.8

Hybrid Local-Cloud QA Pipeline for LangSmith Thread Analysis

Geniable

A hybrid local-cloud QA pipeline for analyzing LangSmith conversation threads, running evaluations, and creating issue tickets.

Components

Component	Description	Documentation
CLI	Local command-line interface for analysis	cli/README.md
Integration Service	AWS Lambda for LangSmith/Jira/Notion integration	cloud/
Evaluation Service	AWS Lambda for running evaluation tools	cloud/

Quick Start (CLI)

# Install
pip install -r requirements.txt

# Initialize (configures credentials and syncs to AWS Secrets Manager)
geni init

# Run analysis
geni analyze-latest

See cli/README.md for full CLI documentation.

---

Lambda Deployment (Legacy)

AWS Lambda function that polls LangSmith annotation queues, analyzes conversation threads for issues, and creates issue pages in Notion.

Features

• On-demand invocation via API Gateway or direct Lambda invoke
• DynamoDB state persistence for tracking processed threads
• Direct Notion SDK integration with dynamic property detection
• Automatic issue classification based on error patterns, performance, and token usage
• SAM-based deployment for infrastructure as code

Architecture

LangSmith Annotation Queue
         │
         ▼
    AWS Lambda (geniable)
         │
         ├──► Analyze threads for issues
         │    - Performance (long execution time > 30s)
         │    - Token usage (high token count > 50K)
         │    - Errors in execution steps
         │
         ├──► Create issues in Notion
         │
         └──► Track state in DynamoDB

Quick Start

Prerequisites

• Python 3.11+
• AWS CLI configured
• AWS SAM CLI installed
• Docker (for building with native dependencies)
• LangSmith API key
• Notion integration token with database access

Configuration

1. Copy environment template:

cp .env.example .env

2. Fill in your credentials in .env:

Variable	Required	Description
LANGSMITH_API_KEY	Yes	LangSmith API key from https://smith.langchain.com/settings
LANGSMITH_PROJECT	No	Project name (default: insights-agent-v2)
LANGSMITH_QUEUE	Yes	Annotation queue name (exact match required)
NOTION_API_KEY	Yes	Notion integration token
NOTION_DATABASE_ID	Yes	Target Notion database ID
NOTION_DATA_SOURCE_ID	Yes	Data source ID for page creation

Deployment

# Build with Docker (required for native dependencies)
sam build --use-container

# Deploy to dev environment
make deploy-dev

After deployment, update Lambda environment variables directly:

aws lambda update-function-configuration \
  --function-name geniable-dev \
  --region ap-southeast-2 \
  --environment "Variables={
    LANGSMITH_API_KEY=your_key,
    LANGSMITH_PROJECT=insights-agent-v2,
    LANGSMITH_QUEUE=Your Queue Name,
    NOTION_API_KEY=your_notion_key,
    NOTION_DATABASE_ID=your_db_id,
    NOTION_DATA_SOURCE_ID=your_ds_id,
    DYNAMODB_TABLE_NAME=langsmith-thread-state-dev,
    AWS_REGION_NAME=ap-southeast-2,
    LOG_LEVEL=INFO
  }"

Usage

Actions

Action	Description
poll	Check annotation queue for new threads
full	Poll + process all new threads (creates Notion issues)
status	Get processing statistics
process	Process a specific thread by ID

Direct Lambda Invocation (Recommended)

Poll for new threads:

aws lambda invoke \
  --function-name geniable-dev \
  --payload '{"action":"poll"}' \
  --cli-binary-format raw-in-base64-out \
  --region ap-southeast-2 \
  /dev/stdout

Full processing (poll + create issues):

aws lambda invoke \
  --function-name geniable-dev \
  --payload '{"action":"full"}' \
  --cli-binary-format raw-in-base64-out \
  --region ap-southeast-2 \
  /dev/stdout

Check status:

aws lambda invoke \
  --function-name geniable-dev \
  --payload '{"action":"status"}' \
  --cli-binary-format raw-in-base64-out \
  --region ap-southeast-2 \
  /dev/stdout

Process specific thread:

aws lambda invoke \
  --function-name geniable-dev \
  --payload '{"action":"process","thread_id":"your-thread-uuid"}' \
  --cli-binary-format raw-in-base64-out \
  --region ap-southeast-2 \
  /dev/stdout

Makefile Commands

make invoke-dev-full    # Poll + process all new threads
make invoke-dev-poll    # Just poll for new threads
make invoke-dev-status  # Get processing statistics
make logs-dev           # Tail Lambda logs
make outputs-dev        # Show stack outputs

API Gateway (requires API key)

# Poll
curl -X POST https://{api-id}.execute-api.ap-southeast-2.amazonaws.com/dev/analyze \
  -H "x-api-key: {api-key}" \
  -H "Content-Type: application/json" \
  -d '{"action": "poll"}'

# Full processing
curl -X POST https://{api-id}.execute-api.ap-southeast-2.amazonaws.com/dev/analyze \
  -H "x-api-key: {api-key}" \
  -H "Content-Type: application/json" \
  -d '{"action": "full"}'

# Status
curl https://{api-id}.execute-api.ap-southeast-2.amazonaws.com/dev/status \
  -H "x-api-key: {api-key}"

Get API key: make get-api-key-dev

Issue Detection

The analyzer identifies these issue types:

Issue Type	Trigger	Priority
Long Execution	Duration > 30 seconds	Medium
High Token Usage	Tokens > 50,000	Medium
Step Errors	Any step with error	Based on severity
Thread Errors	Thread-level errors	Based on severity

Notion Database

The client automatically detects your database schema and adapts to available properties.

Supported properties:

• Title property (required, auto-detected)
• Priority (select)
• Category (select)
• Complexity (select)
• Status (status or select)

AWS Resources Created

Resource	Name
Lambda Function	geniable-{env}
DynamoDB Table	langsmith-thread-state-{env}
API Gateway	geniable-api-{env}
CloudWatch Logs	/aws/lambda/geniable-{env}
CloudWatch Alarm	Error threshold monitoring

Project Structure

langsmith-lambda/
├── src/
│   ├── handler.py              # Lambda entry point
│   ├── config.py               # Environment configuration
│   ├── langsmith_client.py     # LangSmith API client
│   ├── notion_issue_client.py  # Notion SDK integration
│   ├── state_manager.py        # DynamoDB state operations
│   ├── issue_classifier.py     # Issue classification logic
│   └── models/                 # Data models
├── template.yaml               # SAM template
├── samconfig.toml              # SAM configuration
├── Makefile                    # Build/deploy commands
├── requirements.txt            # Python dependencies
└── .env                        # Environment variables (not committed)

Troubleshooting

Check Lambda Logs

make logs-dev
# or
aws logs tail /aws/lambda/geniable-dev --since 10m --region ap-southeast-2

Reset Processed State

To reprocess a thread, delete it from DynamoDB:

aws dynamodb delete-item \
  --table-name langsmith-thread-state-dev \
  --region ap-southeast-2 \
  --key '{"pk": {"S": "THREAD#your-thread-id"}, "sk": {"S": "METADATA"}}'

Common Issues

1. Missing environment variables: SAM parameter overrides may not work correctly with special characters. Update Lambda configuration directly via AWS CLI or console.

2. Notion property errors: The client auto-detects database properties. Check logs to see detected properties: Database properties: [...]

3. Queue not found: Verify the exact queue name matches (including any typos in the original name).

4. No new threads found: Threads are tracked in DynamoDB. Use the reset command above to reprocess.

Development

# Install dependencies
make install-dev

# Run tests
make test

# Run with coverage
make test-coverage

# Format code
make format

# Lint
make lint

# Local testing with SAM
make local

Estimated Costs

Usage	Lambda	DynamoDB	API Gateway	Total
10 invocations/day	$0.50	$1.00	$0.50	~$2/month
100 invocations/day	$2.00	$5.00	$3.50	~$10/month

Cleanup

# Delete dev stack
make delete-dev