REGISTER · CURRENT

Work · shipped systems

Depth over volume.

Sixteen systems, told properly: what broke, what we built, what changed. Every number traces to shipped work; most clients anonymized.

16 RECORDS · 05 NAMED · 11 ANONYMIZED

AI AgentsAI-Powered AppsConversational VoiceDocument IntelligenceData AnalyticsRisk FraudContent CreativeSystem IntegrationsEvaluation Reliability
FIELD · SIXTEEN SYSTEMS ACROSS THE PRACTICE MAP
LEDGER · OUTCOME FIRST, STORY BEHIND EACH ROW
  1. Financial services & markets

    06 SYSTEMS
  2. Citibank

    Fraud decisioning at scale

    A fraud call in milliseconds on every card transaction, five years in production.

    16Mfraud decisions / month
  3. Drip Capital

    Risk portfolio & transaction policy

    A $150M risk portfolio and the policy governing $1B in annual transactions.

    $1Bannual transaction policy · $150M portfolio
  4. A wealth-management fintech

    Millions of securities, one grounded answer

    One grounded answer across four market-data sources and millions of securities, at 95%+ statement-extraction accuracy.

    95%+statement extraction accuracy
  5. A private-markets data platform

    Reading private-markets statements at scale

    150+ inconsistent fund PDF statements from major VC and PE firms, read automatically instead of by hand.

    150+fund statements processed
  6. Two North American private-equity firms, across $10M-$50M portfolio companies

    A data spine for mid-market portfolios

    A shared data foundation for two mid-market private-equity portfolios, lifting operational and marketing efficiency 15-20%.

    15-20%efficiency gains from automation
  7. A retail algorithmic-trading platform

    A visual strategy builder that backtests before it trades

    A node-based builder that designs and backtests trading strategies against market data before any live order.

    Design → backtest → livebefore any live order
  8. Healthcare & life sciences

    02 SYSTEMS
  9. A clinical-intelligence company

    Raw scans to a clinical report

    A pipeline that turns raw medical PDFs into a formatted clinical report, standardizing 100,000+ clinical codes with a clinician's sign-off.

    100,000+clinical codes standardized
  10. A life-sciences commercial-analytics firm serving three top-10 pharmaceutical companies

    Commercial analytics for top-tier pharma

    Marketing-mix modeling and fair territory alignment for three top-10 pharma companies, at 10x the reporting efficiency.

    10×reporting-efficiency improvement
  11. Data & intelligence

    04 SYSTEMS
  12. Newsweek

    Document intelligence at newsroom scale

    A 30M-document search index and 15 production AI apps: retrieval at real scale.

    30Mdocuments indexed
  13. A market-intelligence startup

    A week of analysis in under an hour

    A 7-stage pipeline that turns a week of industry analysis over 100,000+ keywords into under an hour.

    ~100×vs. manual analysis
  14. CGIAR

    Grading nine thousand research records

    9,166 research records graded for real-world impact across five areas by six purpose-fit models.

    9,166research records classified
  15. A Fortune-tier technology company's public-policy team

    Decision support for a public-policy team

    Over a year embedded, turning multi-source program data into decision-support views leaders could steer with.

    1+ yrembedded with the team
  16. AI platforms & tooling

    04 SYSTEMS
  17. Ad Platform (in-house)

    AI video ad production pipeline

    A 13-state production pipeline that scripts, generates, and voices video ads at 5× the output.

    weekly ad output
  18. A generative-AI automation startup

    One pipeline for text, audio, image, and video

    One multi-agent pipeline that turns text, audio, image, and video into controlled outputs, reliable over long runs.

    4modalities in one pipeline
  19. A smart-contract developer-tools company

    Generated code, audited before it ships

    A planner-coder-reviewer-fixer pipeline that machine-audits generated contracts before release, at ~60% less developer effort.

    ~60%less developer effort
  20. An AI reasoning-infrastructure company

    An evaluation harness that catches the model out

    Three public benchmarks run in parallel with a four-layer verifier that catches hallucinations, bad logic, and invalid tool use.

    3public harnesses in parallel
  21. YOUR SYSTEM

    Open a record.

    Bring us the use case. We'll tell you where AI actually helps, and what it takes to ship.

    record pending