2026Agentic AI for maritime business development
VesselReach
An autonomous business-development agent for the shipping industry. It watches incoming vessel traffic, enriches decision-maker contacts, runs outreach and follow-ups, then hands qualified opportunities to a human — because in shipping, hours decide the deal.
Status · Live product — vesselreach.si. Built and operated by me at CatalystIQ.
The problem
In shipping, commercial opportunity has a shelf life measured in hours. A vessel's arrival window, its agent and the buying party are all knowable in advance, but the work of noticing, researching and reaching the right person is manual, repetitive and usually too slow to matter. Business development teams end up contacting ships that have already been supplied.
The approach
- 01Watch vessel movement and port-call signals continuously instead of checking manually.
- 02Resolve each call to the companies behind it, then enrich decision-maker contacts with verification and a confidence level — an unverified address is treated as a weaker lead, not a fact.
- 03Rank the resulting opportunities by fit and timing so the queue is ordered by what is winnable now.
- 04Run first-touch outreach and follow-up sequences automatically, then hand the thread to a human the moment a reply shows real intent.
What I did
- Designed the whole system: ingestion, enrichment, scoring, outreach and the human handoff boundary.
- Built the data model and the enrichment pipeline, including verification and confidence scoring.
- Implemented the agent layer and the guardrails that stop it contacting the same company twice or acting on low-confidence data.
- Shipped and operate the production deployment.
Architecture
Tap any component to see what it does and what it runs on.
Ingestion
Processing
Models
Orchestration
Storage
Interfaces
Select any component to inspect it. Architecture as implemented; component names generalised where they are commercially sensitive.
Trade-offs I would defend
Agent autonomy stops at intent
The agent researches, writes and follows up, but a human owns the conversation once a reply carries commercial intent. Fully autonomous negotiation would be technically possible and commercially reckless.
Confidence over coverage
The system would surface far more contacts if unverified records were treated as valid. Precision is worth more than volume here, because a bounced or wrong-person email costs reputation in a small industry.
Stack & evidence
- Python
- LLM agents
- Postgres
- Scrapers
- FastAPI
- Next.js
- Live site
- vesselreach.si
- Role
- Founder / engineer — sole builder
- Domain
- Maritime commercial operations
Next case study
Fidras
AI ecommerce intelligence platform