Sombra AI · Austin, Texas · English / Español
AI for systems where the numbers have to be right.
Bilingual AI and data systems, in English and Spanish, for crypto protocols and oil and gas operators. Sombra AI builds the layer under the dashboard where the numbers come from.
Scroll ↓
Intelligence is now a commodity.
Everyone has the same models. Anyone can turn them on. Having intelligence is no longer the advantage.
The edge is data shaped so the intelligence pulls through.
Where it comes from. Whether it's clean. How it's structured so a model can reason over it. Who acts on the answer, and how fast.
That's the part Sombra AI builds.
01
Messy source data
Chains, meters, spreadsheets, APIs that disagree.
02
Checked, modeled, scored
Pipelines, quality checks, risk models, AI agents.
03
Decisions people trust
Alerts, scores, and answers that cite their source.
From blockchain protocols to wellheads.
01 · Blockchain & DeFi
Vault indexing across chains, curator attribution, risk scoring, outflow monitoring, and the APIs and docs partners build on.
02 · Oil & gas
Well-test analytics, multiphase meter data quality, anomaly detection, and production trends. Built in English and Spanish.
03 · Energy infrastructure
Data architecture and AI strategy for operators with physical assets. Private AI when the data can't leave the building.
Case 01 · DeFi · In-house product
CuratorWatch: who's running the money in DeFi vaults.
Six pipelines pull every vault into one view, keyed on its on-chain address so nothing gets counted twice. Profiles, comparisons, statistical alerts, and a daily digest.
Case 02 · DeFi · Risk intelligence
An intelligence layer for DeFi risk.
Sombra built the risk scores behind DeFi vault decisions: which vaults are safe, which curators carry risk, and where capital is starting to leave.
01 · Safety score
A 0–100 score across 10 dimensions, including who controls the admin keys. Backtested against 71 vaults that drained to zero: AUC 0.76–0.83, cross-validated.
02 · Default and loss model
A hierarchical Bayesian model of each curator's probability of default, loss given default, and expected loss, built on 3,300+ vault-months.
03 · Outflow monitor
Hourly checks on netflow, momentum, and share-price loss, scored against each vault's own 90-day history and a peer pool.
Case 03 · Oil & gas · Command center
A command center for a $100M+ oilfield services company.
Sombra built the executive command center for a multiphase well-testing company. Live meter data, well-test results, and engineers' shift reports land in one view, in English and Spanish, for the product and engineering team and the executives who run the business.
Case 04 · Oil & gas · Platform
20 years of well tests, one question away.
Sombra modeled wells, tests, and meters as one connected system. Ask in English or Spanish; every answer cites the record behind it. A model router keeps the data on the company's own servers when it has to stay there.
Case 05 · Oil & gas · Field
A meter froze for 16 hours. Nothing flagged it.
Now every channel on the live meter fleet is checked for frozen, stuck, and invalid readings. On top of that: a six-state meter-quality classifier and a stabilization model (held-out AUC 0.79).
Before
+0hdensity channel · reading normal
+0hdensity channel · value stops changing
+4hstill flat · no alarm
+16hstill flat · found by hand
After
+0hdensity channel · value stops changing
+0hfrozen-channel check · FLAGGED
+0haffected well tests marked suspect
How Sombra AI works.
- AuditRead the data and watch the operation before proposing anything.
- PlanA written plan you approve, tradeoffs named.
- BuildThin working slices you can use from week one.
- ProveBacktests, audit trails, numbers you can check.
Each step earns the next.
Got a hard system?
Share what's broken. Sombra AI will say plainly whether it can fix it.
austin@sombraproject.comSombra AI · Austin, Texas · Back to Sombra