Flagship · Aviation
Firestorm Flight Services
An agentic context engine for aviation and trip support. It pulls live FAA and NOAA data into a durable store and distills it into ranked, decision-ready context — deterministic data for flight dispatchers, and structured grounding for the AI agents built on top of it.
Domain
Aviation dispatch & trip support
Shape
A suite of products, built in-house
Status
In active development
Overview
One context layer, two consumers.
Preflight planning runs on data that is scattered, noisy, and unforgiving. A dispatcher has to pull NOTAMs, weather, traffic, and airspace from a dozen sources and reason about what actually matters for one flight, on one day, into one airport.
Firestorm sits underneath that work. It ingests the raw feeds, keeps a durable record, and exposes the same context two ways: as deterministic, queryable data a dispatcher or downstream system can trust, and as structured grounding that keeps AI agents tied to the facts instead of improvising.
Why it’s built this way
- Deterministic first. Safety-critical data is parsed and stored, not summarised away. The numbers are auditable.
- Agents on top, not in the middle. The AI layer reasons over structured context; it never becomes the system of record.
- Durable by default. Fault-tolerant pipelines, replicated storage, and observability wired through from the start.
The suite
Five products, one pipeline.
Each piece stands on its own and hands clean data to the next. Together they take a raw regulatory feed and turn it into a briefing someone can act on.
01
Data pipeline
Ingests live FAA and NOAA feeds — NOTAMs, weather, traffic, airspace, and airport catalogs — into a durable, queryable store. Fault-tolerant pipelines with no single points of failure.
02
Briefing engine
Parses ICAO flight plans into structured flight context, decodes METAR / TAF / PIREP / SIGMET, and groups and ranks NOTAMs by airport and severity.
03
Agent layer
Typed AI workflows that turn structured context into briefing prose and handler communications — grounded in the data, not free-associating from a prompt.
04
Domain knowledge
Codified ICAO and AIP interpretation, country-specific flight rules, and regulatory variance across 40+ countries, so the agents apply consistent judgement everywhere.
05
Briefing surface
Flight context, weather, and NOTAMs presented in pilot- and dispatcher-ready form — the decision-support layer the whole pipeline feeds.
Data sources
Straight from the authorities.
Firestorm reads the same feeds the regulators publish — no scraping, no second-hand aggregators in the safety path.
NOTAMs
FAA NMS API · SWIM (AMQP)
Traffic
TFMS (Solace)
Weather
NOAA — METAR / TAF / SIGMET / AIRMET
Airports & airspace
NASR · CIFP · OurAirports
Under the hood
Boring where it counts.
Small binaries, durable formats, and replication you can reason about. The clever part is the domain modelling, not the infrastructure.
Language
Go
Streaming
NATS · AMQP · Solace
Orchestration
DagNats workflow engine
Storage
SQLite · Litestream to R2
AI
Claude · typed workflows
Observability
OpenTelemetry to SigNoz
Deploy
Single binaries · ADR-driven
Status
In active development
Working in aviation or trip support?
If Firestorm is relevant to what you’re building or buying, I’m happy to talk through where it is and where it’s headed.