Enquiry Email Automation Agent
Ops stopped retyping freight quote emails — mailbox messages become Shipsoft CRM enquiries at 97% accuracy, saving about 10 minutes per enquiry.
Role: AI Engineer · Jan 2026 – Present · internal tooling (no public demo)
IMAP → CrewAI → Shipsoft
Problem
Freight enquiries arrive as messy emails—carton lines, ports, contacts, FCL/LCL/AIR cues—while ops still need clean data in a multi-step enquiry form. Manual re-entry is slow and inconsistent. Enquiry Agent watches a mailbox, extracts structured fields with CrewAI/OpenAI, and optionally creates the enquiry in Shipsoft.
Role
AI Engineer on the Shipsoft ecosystem at Shipsoft Solutions. Designed and built the enquiry automation pipeline end to end—IMAP ingest, Celery/Redis workers, CrewAI extraction, Postgres job state, Shipsoft CRM delivery, operator docs, and runbooks. (Internal deployment only.)
What I built
- End-to-end architecture, env setup, and operator runbook for the enquiry automation pipeline.
- Celery/Redis topology with Postgres dual-write for job and mailbox state, plus Shipsoft login–impersonate–create–leave delivery.
- IMAP IDLE ingest with resume/catch-up so mail isn’t missed across reconnects.
- Async extraction workers with profile-aware CrewAI prompts for LCL carton and general freight emails.
- Deterministic finalize step and API payload shaping aligned to Shipsoft’s enquiry create contract.
- Optional OpenAI vision path for cargo details in image attachments.
- Durable job tracking and per-job / Shipsoft HTTP logging with credential redaction.
- Pytest coverage for preprocess, finalize, models, and API payload shaping.
Stack
Python, Celery, Redis, PostgreSQL, SQLAlchemy, CrewAI, OpenAI, Pydantic, IMAP, Shipsoft tenant API.
Related
Part of the Shipsoft 2.0 freight CRM platform — this agent feeds the enquiry create API rather than replacing the multi-tenant SaaS itself. Sister pipeline: Rate Sheet Agent (rate sheets → quoting JSON).