ExxonMobil

Real-time intelligence at global scale.
Streaming data, safety reporting, and executive analytics across a highly distributed campus estate.
- 58K
- employees · year-end 2025
- 80+
- campuses supported
- 400+
- employees reached
Trusted by teams at Explore the work

Streaming data, safety reporting, and executive analytics across a highly distributed campus estate.
Configuration-driven GCP pipelines and precisely branded analytics that product managers could launch without code changes.
A high-availability Tableau rollout, consolidated data model, and executive pricing dashboards delivered without disruption.

A high-availability Tableau migration, automation, and hands-on engineering enablement completed three times faster than expected.
A complete analytics delivery system spanning production SQL, reusable Tableau products, and Python automation.
Pipelines and dashboards re-architected for trillion-record scale across major US mobile carriers.
From Little Ideas To Space
Software development and engineering in Austin, Texas: custom apps, data, and AI for teams that move fast.
Behind the game below
We built this explainer 10 times: 7 versions with Claude Code and 3 with Codex. The one below is Claude #5. Every one of them is a click away.
Claude Code · 7 versions, built in this repo as Astro + three.js components
An exploding mechanical watch movement: mainplate = diagram, jewels = interviews, mainspring = voice, gear train = backlog, escapement = agents, dial = systems. Shelved: the metaphor needed translating and the scene was cluttered.
Open full page (opens in a new tab)
A real-time-strategy base builder on a fixed isometric board: dev3lop is the HQ. The dev3lop cursor scouts your systems, interviews your people, turns speech into a build queue and sends 4,096 agents to build it; every cloud, database and API runs a supply line to HQ; then the map is replayed with four vendors and an illustrative J-curve shows profit growing. Judges ranked the pitch #1.
Open full page (opens in a new tab)
A calm, flat Mini Metro-style transit map: dev3lop is the one operator. Stations are your systems and people, dashed track is the plan and solid track is built. Interviews meet at one broken handoff, transcripts become 12 numbered tickets, 3,072 agents lay the track, every cloud, database, API and app line runs through one hub, then the map breaks into four vendors and reconnects.
Open full page (opens in a new tab)
An isometric tycoon campaign in 10 levels: each level unlocks one building for one step of the process and a named dev3lop cursor plays it, from mapping your systems to drag-selecting 2,048 agents and piping in every resource. Hard mode splits the board between four vendor cursors so profit leaks, then an illustrative year fast-forwards.
Open full page (opens in a new tab)
A Shapez/Factorio-style conveyor factory with a quest log and item hotbar: voice becomes transcripts, transcripts become tickets traced to a person, 2,048 agent bots stream from the tower through the rack into the build bay, and your clouds and databases plug into one line. Then the same factory splits between four vendors and stalls until the glue goes back on.
Open full page (opens in a new tab)
Its 9-step story is the game just below this section. Go to the game
A quest-driven RPG overworld of your company: the dev3lop party (six roles, one team) maps the towns, interviews the people and connects their quotes, types out the transcript under the voice, posts STORY-214, and sends 2,048 agents across the map. Then the same map is replayed as separate vendors (profit leaks at 4 handoffs) and as one team (gold roads, a J-curve).
Open full page (opens in a new tab)
The judge panel's synthesis: the RTS base with the best grafts from every other pitch — a named dev3lop cursor, one problem thread (the re-keyed CRM export) through every level, inbound vs outbound trunks, and a Glue ON/OFF profit toggle.
Open full page (opens in a new tab)
Codex · 3 versions, built outside the repo as standalone three.js pages
Codex iteration 1: an exploded precision watch you can drag and assemble, with a guided six-phase story (Discover, Listen, Translate, Orchestrate, Deliver, Compound) and the cloud, data and stack names streaming into one shared context.
Open full page (opens in a new tab)
Codex iteration 2: every step opens its own animated 3D process scene lifted out of the watch — interviews converging on a shared requirement, a typed transcript with NLP extraction into backlog item INT-014, four clusters of 512 agents with a human release gate, systems converging on dev3lop, and a spatial growth chart you can disconnect.
Open full page (opens in a new tab)
Tells the same 6-chapter story as Codex C1; what changes is how it looks and plays.
Codex iteration 7, its latest: adds a plain-language guide for all 32 buildings (what each does, when it helps, a concrete example) and opens in exploration mode with an opt-in tour.
Open full page (opens in a new tab)
Play through how we turn your people’s conversations into connected software, and why keeping every step on one line is what pays off.
How we build
Every build we run works like a production line: your people’s conversations go in, working software comes out, and profit is the point. Scroll to watch the line get built, one station at a time.
Step 1 · Diagram
We diagram how work moves through your CRM, spreadsheets and billing today, name the target, pin the milestones and gather the requirements, before any code is written.
Step 2 · Interviews
Requirements live in people’s heads, so we interview the ones who do the work. Then we connect the dots: when finance and ops describe the same spreadsheet from opposite ends, that link becomes a requirement.
Step 3 · Transcription
Each conversation is recorded and transcribed. Natural-language processing tags the systems, pain points, deadlines and asks, so nothing depends on someone’s notes.
“Every Monday I export the CRM to Excel and re-key it into billing before month-end.” Example · Interview 3 · Ops lead · 00:14:32
Step 4 · Backlog
Tagged phrases become epics, stories and acceptance criteria. Every ticket traces back to the person and the minute it came from, so the backlog is what your people asked for.
Step 5 · AI agents
dev3lop spawns thousands of AI agents against the backlog. Each one claims a ticket, builds it, tests it and hands it back. Our engineers set the priorities and approve what ships.
Each dot in the scene is one agent. The tower is the one coordinator; green means tested and handed back.
Step 6 · One company
Your clouds, databases and APIs plug straight into the line, in the cloud or on your own servers. Out come web apps and desktop software, built end to end.
Without the glue
One vendor for discovery, one tool for notes, another for the backlog, a separate shop for the build. The same work piles up at every handoff, context leaks out, and profit stalls.
With the glue
Put the line back together and work flows end to end. After a short build-up, the time saved at every handoff adds up month over month, and so does margin.
The chart is illustrative, not client data.
This game is 1 of 10 versions of the same story. See every version ↑
Add frontend, backend, and design to your toolbelt.
Pipelines · Warehouses · Infra
HoverEvent streams, Airflow DAGs, Snowflake/BigQuery/Redshift warehouses. We build the pipes that never leak.
Explore serviceTableau · Power BI · Looker
HoverDashboards executives actually open. Modeling, semantic layers, and the metrics that drive the room.
Explore serviceAWS · Azure · GCP
HoverMigrations, cost optimization, infrastructure-as-code. Pay less. Ship faster. Sleep better.
Explore serviceAPIs · Full-stack · Data products
HoverCustom apps that wrap your data in a product. React, Node, TypeScript, and the taste to know when to stop.
Explore serviceStar schemas, dbt, semantic layers
HoverClean dimensional models and dbt transformations that make analytics 10x cheaper to run and 100x easier to trust.
Airflow · Dagster · Prefect
HoverDAGs that retry, alert, and backfill themselves. We write the playbook so your on-call doesn't have to.
Agents · LLMs · RAG
HoverProduction AI agents on your data. Ollama, OpenAI, Claude. We pick the right model for the right cost.
Access control · PII · SOC 2
HoverRow-level security, data masking, audit logs. The boring stuff that keeps you out of the news.
Query optimization, indexing
HoverDashboards that load in 300ms instead of 30 seconds. Warehouse bills cut in half. Math, not magic.
Great Expectations · Monte Carlo
HoverTests, observability, and anomaly detection so you catch bad data before the CEO does.
Pure CSS. Zero JavaScript. Server-rendered. Fast by design.
Software · API-first team messaging
HoverDeploy chat on your own infrastructure with REST + SSE APIs and pricing that tracks servers, not seats.
Explore productSoftware · Visual ETL
HoverBuild asynchronous pipelines through a visual DAG interface for transforms, joins, and production data outputs.
Explore productSoftware · Job automation
HoverOrchestrate recurring tasks, chained dependencies, and failure alerting with a lightweight scheduler.
Explore productSoftware · Intelligent agents
HoverRun autonomous workflows with business-context-aware agents and human checkpoints when needed.
Explore productSoftware · Blogging distribution
HoverShip content with model-assisted drafting, interlinking, channel recommendations, and publishing validation.
Explore productSoftware · Managed SEO agent
HoverA managed Bee stack with premium OpenAI-backed generation, image support, and hosted infrastructure.
Explore productSoftware · Offline LLM editor
HoverCode editor designed for local and offline LLM workflows with compliance-forward deployment control.
Explore productSoftware · SEO intelligence
HoverUnify rank history, audits, competitor checks, backlog tracking, and action-focused SEO strategy.
Explore productPlatforms and products we're creating.
Collaborative custom platforms
HoverDesign and ship bespoke digital products with collaborative delivery focused on measurable business outcomes.
Visit siteAI consulting agency
HoverOperationalize AI initiatives with practical architecture, model strategy, and implementation support for production teams.
Visit siteQuick guide
No AI. No database. Nothing is saved here.