Agents alreadyrunning on Mondays.
First card is the active Raiagents, LLC pilot. The rest are production AI / ML systems shipped before raiagents was founded, delivered while at gAI Ventures, Routesense, and Amazon, plus a solo-founded consumer app. Numbers are real and freshly measured. Reach out for unredacted references on request.
- NX-001 · NaturExchangeRegulated compliance · 2026 · active
Six AI co-pilots that give every compliance practitioner the leverage of a research team.
An agentic-AI platform for a global compliance firm whose practitioners file across more than a hundred national regulatory regimes. Six specialist co-pilots run alongside each practitioner: one auditing risk, one drafting submissions in the form each authority expects, one watching for regulatory changes, one tracing supply-chain provenance, one matching marketplace deals, and one defending patents and digital sequence claims. The agents take on the retrieval, drafting, and watching that used to swallow most of a practitioner's week. The practitioner reviews, edits, and signs every output. Every claim traces back to a primary source with a citation and retrieval date. Built as a 90-day phased pilot with hard review gates. The next phase budget releases only if the previous gate passes.
First demo shipped
28 days
Output per practitioner
5–10× more
Regimes in scope
10 → 142
Multi-agent orchestrationRAG over a regulatory corpusClaude Sonnet + HaikuPostgresVector searchHuman-in-the-loop review queueReproducible audit trailRaiagents, LLC · current pilot - C-001 · UNO Digital BankBanking · 2026
An AI analytics platform that lets bank staff ask their data questions in plain English.
A six-agent system that lets non-technical staff at UNO Digital Bank query the bank's customer data without writing a line of SQL. Behind the scenes, agents take a question apart, find the right tables, run the query, and explain the answer in plain language with sources. The system was grounded against 262 reference question-and-answer pairs the bank's analysts curated, so it rarely guesses. Shipped end to end in 10 weeks, including security review, full audit trail, and zero incidents where the system modified data it shouldn't have.
Time to production
10 weeks
Response time
40% faster
Data-safety incidents
0
AWS BedrockClaudePostgresFastAPIMulti-agent orchestrationAI Product Engineer · gAI Ventures - C-002 · RoutesensePayments · 2025
A real-time routing brain that picks the right payment processor for each transaction.
The decision engine at the heart of Routesense's payments infrastructure. Every incoming transaction has several possible payment processors that could handle it; the engine picks the right one in real time, based on each processor's recent approval rate, current health, and risk profile. Built a scoring system that rates each processor dynamically, plus failover and adaptive load-balancing logic so the system stays up if any one processor degrades. Took the company's payment infrastructure from zero to production.
Transactions approved
+8–10%
Processing cost
2–5% lower
Build
Zero to production
Real-time decisioningHealth scoringFailover & adaptive load balancingDistributed servicesAI Engineer · Routesense - C-003 · Internal · SwikFintech · 2026
An invoice-factoring platform that compresses borrower verification from days to hours.
Designed and built Swik 2.0 end to end: data model, backend, frontend, and the full AI verification pipeline. The original process (calling each prospective borrower's customers to confirm invoices were real) took human reviewers days per applicant. The new system handles it through a coordinated set of agents: one drafts personalized verification emails, one places AI phone calls, one navigates customer portals using vision models, and a multi-stage extraction chain pulls the signed invoice data out of PDFs.
Verification time
days → hours
AI services in pipeline
12
Build time
6 weeks
FastAPINext.jsPostgresAI phone callsVision LLMsMulti-stage PDF extractionAI Product Engineer · gAI Ventures - C-004 · Internal · SwikKnowledge / RAG · 2026
Unified search across a company's email, drives, and chat, fresh to the second.
Built Swik AI, a search-and-retrieval system that pulls a team's Gmail, Google Drive, Outlook, OneDrive, and Dropbox into one searchable index. Most enterprise search systems re-index overnight; this one updates within seconds of a new file or email arriving, through real-time webhooks and incremental sync. Each user only sees what they're allowed to see. Powers downstream agents that need to ground answers in your company's actual documents, with citations and a per-query audit trail.
Apps unified
5
Index freshness
~1 second
Permission model
Per-user
Vector searchWebhooks + incremental syncFastAPIPostgresOAuthAI Product Engineer · gAI Ventures - C-005 · Internal · SwikF&B / Risk ML · 2025
Flagging restaurants in operational distress 60–90 days before the financials show it.
A risk-prediction system that combines public review data, health-inspection records, and statistical change-point signals to predict which California restaurants are sliding into operational distress, well before the lagging financial indicators catch up. Distributed crawlers ingested over a million Yelp and Google reviews plus health-inspection records from 40,000+ establishments. The model flags 87% of true high-risk businesses while cutting false alarms by more than half compared to a simple threshold-based approach.
True high-risk caught
87%
Lead time vs financials
60–90 days
False alarms vs baseline
−55%
Gradient-boosted treesTime-series change-point detectionDistributed web crawlingML Engineer · gAI Ventures - C-006 · AmazonPayments · 2022–2025
Payment-fraud ML at Amazon scale, holding a 99.9% uptime promise across billions of daily transactions.
Built and operated payment-fraud and anomaly-detection systems supporting roughly 30% of Amazon's internal transaction volume. The systems flag suspicious patterns in real time and route transactions accordingly. Refactored the underlying event-driven services to cut both response time and infrastructure cost, while keeping the 99.9% uptime promise. At this scale, every fraction of a percentage point of unavailability translates to meaningful customer impact.
Payment risk reduced
78%
Response time
45% faster
Uptime SLA held
99.9%
AWS LambdaSageMakerS3Event-driven servicesKinesisSoftware Engineer, ML · Amazon - C-007 · GamificationFitness LLCConsumer · Health · 2024–
Cerro: a fitness coach in your pocket, powered by Apple Watch data and a personal AI agent.
Solo-founded Cerro, a fitness coaching app that pairs Apple Watch and HealthKit sensor data with a personal AI coach. The coach has full context on your sleep, training history, heart rate, and recent workouts, so when you ask 'should I push hard tomorrow?', it actually knows. A gamified engagement loop and real-time progress tracking shaped the experience. Beta users were retained at 70% above the baseline.
Beta retention
+70%
Platform
iOS
Sensor data
Apple Watch + HealthKit
SwiftHealthKitFastAPIOpenAIPostgresFounder
Currently booking · 4 slots / mo
Tell us about your business.
We'll tell you what to automate first.
Free 30-minute discovery call. We talk through your business, where AI could actually help, what it would cost, and how long it would take. No pressure to commit. If we're not a fit, we'll say so.