AutoTasker AI – Smart Multi-Agent Task Automator with Cloud Services
PROJECT OVERVIEW: AutoTasker AI
A cloud-based intelligent assistant that understands natural-language commands,
plans a workflow using agents, executes the tasks via AWS Lambda, and stores
results/logs in S3 and DynamoDB.
Concept:
You give natural language commands like:
“Every Monday, generate a summary of my GitHub commits, email it, and save a log.”
The system:
Understands your task (LLM)
Uses agents to plan and execute steps
Saves reports/logs to the cloud database
Schedules & runs via cloud services
Tech Stack:
Gen AI: LLM (Claude, GPT-J, or Bedrock on AWS) to understand instructions
Agentic AI: LangChain or AutoGen to break into steps
AWS Lambda: Serverless task executor
AWS EventBridge / Step Functions: Automation triggers
AWS SES: Send emails
Amazon DynamoDB: Store task history
Amazon S3: Store logs, files
Use Case: O ice automation, virtual productivity assistant
Scalable: Add Slack, Discord, WhatsApp integrations later
Flow of the System
A[User Input (Natural Language)] --> B[LLM Task Interpreter]
B --> C[Agent Planner (LangChain/AutoGen)]
C --> D[Task Scheduler (AWS EventBridge)]
D --> E[Executor (Lambda + API calls)]
E --> F{Store Result}
F --> G[Email Report via SES]
F --> H[Save Logs to S3]
F --> I[Save Metadata to DynamoDB]
MODULE-WISE BREAKDOWN
1. User Input & Task Interpretation
Frontend or CLI lets user give a command like:
“Every Monday at 9 AM, generate a summary of my GitHub commits, email it, and save
to logs.”
This input is sent to an LLM (e.g., Claude, GPT-J, or Bedrock) which:
o Parses the frequency (“Every Monday”)
o Recognizes the API needed (GitHub)
o Recognizes actions: summarize, email, save
Output: Structured plan in JSON
{
"schedule": "cron(0 9 ? * MON *)",
"steps": [
{ "type": "fetch", "source": "GitHub", "action": "getCommits" },
{ "type": "process", "action": "summarize" },
{ "type": "output", "method": "email" },
{ "type": "output", "method": "logToS3" }
]
}
2. Agentic Planning (LangChain / AutoGen)
The structured JSON is fed into an Agent system:
o Splits the task into atomic steps
o Figures out dependencies (e.g., fetch before summarize)
o Can even retry failed steps automatically
o Calls appropriate APIs (GitHub, Polly, etc.)
Agent builds a runnable chain of actions
3. Task Scheduling (AWS EventBridge or Step Functions)
Once the plan is created:
o A schedule is created using AWS EventBridge (Cron-style)
o It will trigger AWS Lambda every Monday at 9 AM
Automation achieved without needing a server running 24/7
4. Task Execution (AWS Lambda)
Lambda functions are small, stateless Python/[Link] scripts:
o Fetch GitHub commits using GitHub API
o Use LLM to summarize text
o Compose HTML report
o Send email via AWS SES
o Log output to S3 (for long-term storage)
o Store metadata (task name, runtime, status) in DynamoDB
Each Lambda is lightweight, secure, and scalable
5. Storage & Logs
Storage Use
S3 Store generated reports (PDF/HTML/logs)
DynamoDB Store metadata like task name, timestamp, result
SES Send email notifications or reports
This makes the system modular, stateless, and fully cloud-native
HIGH-LEVEL WORKFLOW
Example Command:
“Every Monday, summarize my GitHub commits, email me the report, and save it to
logs.”
Full Workflow:
1. User Input
→ via CLI / web (e.g., “Summarize GitHub every Monday”)
2. LLM Prompting
→ Convert to structured JSON: actions, timing, sources
3. Agentic Planning
→ LangChain or AutoGen splits task into steps
4. Task Scheduling
→ AWS EventBridge sets up cron-like triggers
5. Execution (Serverless)
→ Lambda function runs: GitHub fetch → LLM summarize → report → SES email +
S3 save + DB log
6. Logging and Monitoring
→ CloudWatch + DynamoDB store status, outputs
CLOUD INTEGRATION (How services connect)
Module AWS Service
Task scheduler EventBridge
Function execution Lambda
Logs and reports S3
Email delivery SES
Task metadata DynamoDB
Monitoring CloudWatch Logs
LLM backend (optional) Bedrock / OpenAI / Local model
GITHUB FOLDER ARCHITECTURE
autotasker-ai
│
├── frontend/ # (Optional React or Streamlit UI)
│ └── [Link]
│
├── backend/
│ ├── lambda/
│ │ ├── [Link] # Lambda entry point
│ │ ├── github_fetch.py # GitHub API fetcher
│ │ ├── [Link] # LLM summarizer
│ │ ├── email_report.py # SES integration
│ │ ├── s3_storage.py # S3 uploader
│ │ ├── db_log.py # DynamoDB writer
│ │ └── [Link] # Shared utilities
│ │
│ ├── lambda_deploy.zip # Packaged code for Lambda
│ └── [Link]
│
├── agent_planner/
│ ├── prompt_engineering.md # Prompt design for task parsing
│ ├── task_parser.py # LLM-based planner
│ └── langchain_chain.py # Optional LangChain implementation
│
├── scheduler/
│ ├── schedule_creator.py # EventBridge scheduler logic
│ └── [Link] # IAM policy for Lambda access
│
├── data/
│ ├── sample_tasks.json # Example outputs from prompts
│ └── test_github.json
│
├── architecture_diagram.png
├── [Link]
└── setup_guide.md