Skip to content
shamathmikaPublic

About

LangGraph multi-agent orchestrator that plans a move as a typed task checklist and gates every automated portal submission on human approval

Topics

Resources

Stars

0 stars

Watchers

0 watching

Forks

Latest commit

 

History

51 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 
 
 
 
 

Repository files navigation

Transitly - Al powered moving assistant architecture

⭐ Core Mission

Make the paperwork side of moving disappear. You enter your old/new address and move dates; Transitly plans your tasks and, with your approval, autofills the right forms (USPS, utilities, ISP) while keeping an auditable checklist and timeline.

🔧 The Problem We Solve

Moving requires dozens of small but critical admin steps spread across different portals. People forget tasks, mistype information, and lose hours clicking through brittle forms.

  • Fragmented workflows: USPS, utilities, ISP, insurance—every site is different and easy to miss.
  • Manual, error‑prone forms: Re‑typing addresses and dates across portals leads to mistakes and bounces.
  • No single source of truth: Checklists live in notes or memory; progress and proofs are hard to track.
  • Time sinks and deadline risk: Missed windows (e.g., tech appointments) create costly delays.

💡 Our Solution

An agentic system that plans, acts (with approvals), and tracks your move‑day admin.

  • Smart intake → plan: Collect from_address, to_address, move_out_date, move_in_date (plus household prefs) and generate a dynamic checklist tailored to your move.
  • Approve to autofill: Use a browser‑action agent to navigate real websites and prefill up to the review/submit page; a human click is required to confirm.
  • Live status + audit: Stream progress to the UI via SSE, save checklists in DynamoDB, and keep structured action logs. Screenshots/receipts and reminders are planned next.
  • Privacy‑first: No passwords are stored; sensitive fields are masked and PII is redacted in logs.

🔄 Multi-Agent Orchestration System

Transitly uses a LangGraph‑based orchestrator that routes between specialized workers and yields a consistent state: messages, user_details, a typed checklist, and action results.

  • Orchestrator Agent: Supervises the workflow using a reasoning LLM and a registry of workers. Chooses next_task deterministically when possible and stops when all relevant items are complete or blocked.
  • Get User Details Agent: Resolves a user_id and fetches profile + latest move from DynamoDB via a tool. Idempotent: skips if details already exist and updates any matching checklist item.
  • Checklist Agent: Plans the move as ChecklistItem objects, assigning agent_label (e.g., amazon_address_change) so the supervisor can route execution. Produces 4–10 atomic, actionable tasks.
  • Nova Act Agent (Amazon Address Change): Uses NovaAct browser automation to update the default Amazon shipping address. Runs only with user approval, requires login in a real browser, and stops at the review page for a human click. Results are written back to state.

Backend: FastAPI with Cognito JWT auth, DynamoDB tables (TransitlyUsers, TransitlyMoves, TransitlyChecklists), and an SSE endpoint (/run-agents-stream). Frontend: React + Vite; the AgentStatus component listens for text/event-stream updates and renders progress and checklists.

✨ Key Benefits

  • Fewer clicks, fewer mistakes: Autofill reduces re‑typing and address errors across portals.
  • Always know what’s next: A living checklist with statuses, streamed live to the UI.
  • Approval in the loop: Every action is gated by a “Confirm & Autofill” step; no silent submissions.
  • Cloud‑ready architecture: Cognito for auth, DynamoDB for state, Bedrock/Gemini‑class LLMs, and Nova Act for safe browser actions.
  • Extensible workers: Add agents for USPS COA, utilities, ISP transfers, and DMV reminders without changing the orchestration model.

About

LangGraph multi-agent orchestrator that plans a move as a typed task checklist and gates every automated portal submission on human approval

Topics

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages