# Devzilla

> Devzilla is a forward-deployed AI engineering agency. Senior engineers embed directly within client teams and remain accountable from technical strategy through production deployment. The Lab, Foundry, and Signal support that Agency work through applied research, in-house product building, and first-hand publishing.

## Site

- [Home](https://devzilla.co/): Devzilla company overview and the Research, Build, Launch, Repeat operating loop.
- [The Agency](https://devzilla.co/agency/): Forward-deployed AI engineering services, delivery model, and engagement details.
- [The Lab](https://devzilla.co/lab/): Applied AI and product research streams that inform Devzilla client and product work.
- [The Foundry](https://devzilla.co/foundry/): Devzilla in-house product model, venture stages, and current Foundry status.
- [The Signal](https://devzilla.co/media/): Devzilla essays and technical field notes on applied AI, engineering craft, and building products.
- [Contact](https://devzilla.co/contact/): Project inquiry form and direct contact information for Devzilla.
- [About Devzilla](https://devzilla.co/about/): Verified company model, leadership, operating principles, and public work.
- [Devzilla Work](https://devzilla.co/work/): Index of real Devzilla-owned public engineering work and technical case studies.
- [AI Engineering Services](https://devzilla.co/services/): Detailed Devzilla AI engineering capabilities, engagement situations, deliverables, and frequently asked questions.
- [UFO Disclosure Map](https://devzilla.co/work/ufo-disclosure-map/): Devzilla turned a difficult government disclosure archive into an interactive geospatial system with searchable events, documents, imagery, and remote media.
- [Earthquake Watch](https://devzilla.co/work/earthquake-watch/): Devzilla built a publication pipeline over the United States Geological Survey event service, including a California polygon filter that rejects out-of-state events returned by the source bounding box.
- [SF Building Permits](https://devzilla.co/work/sf-building-permits/): Devzilla built a source, scoring, records, and publishing workflow over San Francisco OpenData so notable permits become contextual reports instead of rows in a municipal dataset.
- [Forward-Deployed AI Engineering](https://devzilla.co/services/forward-deployed-ai-engineering/): A forward-deployed AI engineer works inside the client team instead of handing requirements to a remote delivery queue. Devzilla engineers join the working cadence, build in the real stack, own technical decisions through deployment, and leave the client with documented software the client can operate without permanent agency dependence.
- [AI Prototype-to-Production Engineering](https://devzilla.co/services/ai-prototype-to-production/): Devzilla first identifies what the prototype proves and what it still avoids. The production build then adds real data contracts, evaluation, failure handling, security, observability, cost controls, deployment automation, and ownership. The goal is a system that survives normal operations, not a more polished demonstration.
- [Agent and LLM Systems](https://devzilla.co/services/agent-and-llm-systems/): A production agent needs bounded responsibilities, typed tools, explicit permissions, observable state, repeatable evaluation, cost and latency limits, and a safe escalation path. Devzilla treats the model as one component inside that operating system rather than asking an unconstrained prompt to carry the entire product.
- [Knowledge Graph Systems](https://devzilla.co/services/knowledge-graph-systems/): A knowledge graph is useful when the relationships among people, systems, assets, events, and decisions matter as much as the records themselves. Devzilla uses graphs when teams need durable identity, provenance, connected retrieval, explainable paths, or shared memory that cannot be represented reliably as isolated documents and embeddings.
- [Geospatial Systems](https://devzilla.co/services/geospatial-systems/): Geospatial engineering makes location a reliable part of the product model. It covers source coordinates, projections, boundaries, spatial joins, distance, indexing, map interaction, and geographic validation. Devzilla builds the data pipeline and user interface together so the map reflects correct spatial reasoning instead of decorating ordinary records.
- [A Bounding Box Is Not California](https://devzilla.co/media/california-earthquake-filtering/): How Earthquake Watch uses a state polygon to keep neighboring Nevada, Oregon, and Baja California events out of a California publication.
- [The Twelve-Gigabyte Map We Did Not Deploy](https://devzilla.co/media/ufo-map-hybrid-media-architecture/): Why the UFO disclosure map bundles stable images while leaving government documents, video, and audio on their primary hosts.
- [From Permit Rows to a Focused Publication](https://devzilla.co/media/building-permit-publication-pipeline/): The source, records, and rendering decisions behind SF Building Permits, a publication built over San Francisco OpenData.

## Consolidated context

- [Full context](https://devzilla.co/llms-full.txt): Approved company positioning and page summaries in one text file.