VIVIMETRIX PLATFORM
Architect (2021–2023), then CTO & Chief Architect (2024–present)
Real-time fintech SaaS platform with a self-tuning pattern recognition engine.
THE CHALLENGE :
Build a real-time trading platform and the engineering organization behind it, starting from nothing, in a crowded fintech market.
The platform had to read fast-moving market data across ~10,000 stocks every trading minute, reliably recognize complex chart patterns, alert subscribers in under a second, and stay dependable as both the customer base and the team grew.
APPROACH & KEY DECISIONS :
I treated the core problem as perception under noise, not just data processing. The engine had to "see" patterns in raw price data the way an experienced trader does, ignore the noise, and get better over time.
Key decisions:
Designed the pattern recognition engine from first principles instead of relying on existing libraries
Used geometric analysis (triangles and slope relationships) to separate real patterns from noise
Made the engine self-tuning: every day it scored every pattern, alerted and abandoned, against actual outcomes and adjusted its own parameters for the next session
Prioritized performance through in-memory design and parallel processing
Built the engineering team, processes, and culture alongside the product, not after it
SOLUTION & IMPLEMENTATION :
I architected the full platform and led its build while growing the engineering team. That included:
A pattern recognition engine that processes 10,000 charts in under 10 seconds
Real-time data pipelines and alerting with a sub-second delivery SLA
Cloud-native architecture on AWS (EKS) with strong observability
Multi-tenant isolation with row-level security and encryption, redundancy, and disaster recovery
PCI-compliant payments through an event-driven Stripe integration
Team structure, hiring, sprint process, CI/CD, and code quality practices
Through repeated optimization and caching, per-chart analysis went from multiple seconds to 5 ms in production.
RESULTS & IMPACT :
99.99% platform uptime
Pattern detection at 5 ms per chart, with alerts delivered in under 1 second
Architecture that carried 100% of revenue through the first $1M+ and several thousand users
78% average annual revenue growth since 2022; new pattern features lifted subscription revenue ~50% at launch
Engineering grown from 5 to 15 under four managers, with change lead time cut ~40% (14 days to 8 days)
Customer satisfaction raised from 87% to 96%; Sev 0–1 incidents cut from 8 to 3 a year
The self-tuning engine became the core product differentiator
Patent Application
Method, System and Device for Predicting Stock Performance and Building an Alert Model for Such Estimation
US 2022/0084119 A1 · Filed November 17, 2021
Describes a system that analyzes continuous time-series market data to identify and validate bullish "W" and bearish "M" patterns using precise shape and timing rules. It filters noise with multi-stage anomaly detection, forecasts reversal zones and entry and exit levels, and feeds validated patterns into adaptive models that update in near real time and trigger customized alerts.
WHAT I LEARNED :
Building the product and the organization at the same time taught me that engineering quality and team health can't be separated.
Introducing AI-assisted development also surfaced a real fear among engineers of being replaced. Adoption depended less on the tools than on clear expectations, accountability, and a culture that treats AI as a multiplier, not a threat. The biggest gains came from pairing new capabilities with human judgment: every change still gets human review.
TECHNOLOGIES :
C# · .NET · SQL Server · AWS (EKS) · In-memory caching · Parallel processing · Event-driven · Real-time systems · Adaptive algorithms
VIVIMETRIX PLATFORM
Architect (2021–2023), then CTO & Chief Architect (2024–present)
Real-time fintech SaaS platform with a self-tuning pattern recognition engine.
THE CHALLENGE :
Build a real-time trading platform and the engineering organization behind it, starting from nothing, in a crowded fintech market.
The platform had to read fast-moving market data across ~10,000 stocks every trading minute, reliably recognize complex chart patterns, alert subscribers in under a second, and stay dependable as both the customer base and the team grew.
APPROACH & KEY DECISIONS :
I treated the core problem as perception under noise, not just data processing. The engine had to "see" patterns in raw price data the way an experienced trader does, ignore the noise, and get better over time.
Key decisions:
Designed the pattern recognition engine from first principles instead of relying on existing libraries
Used geometric analysis (triangles and slope relationships) to separate real patterns from noise
Made the engine self-tuning: every day it scored every pattern, alerted and abandoned, against actual outcomes and adjusted its own parameters for the next session
Prioritized performance through in-memory design and parallel processing
Built the engineering team, processes, and culture alongside the product, not after it
SOLUTION & IMPLEMENTATION :
I architected the full platform and led its build while growing the engineering team. That included:
A pattern recognition engine that processes 10,000 charts in under 10 seconds
Real-time data pipelines and alerting with a sub-second delivery SLA
Cloud-native architecture on AWS (EKS) with strong observability
Multi-tenant isolation with row-level security and encryption, redundancy, and disaster recovery
PCI-compliant payments through an event-driven Stripe integration
Team structure, hiring, sprint process, CI/CD, and code quality practices
Through repeated optimization and caching, per-chart analysis went from multiple seconds to 5 ms in production.
RESULTS & IMPACT :
99.99% platform uptime
Pattern detection at 5 ms per chart, with alerts delivered in under 1 second
Architecture that carried 100% of revenue through the first $1M+ and several thousand users
78% average annual revenue growth since 2022; new pattern features lifted subscription revenue ~50% at launch
Engineering grown from 5 to 15 under four managers, with change lead time cut ~40% (14 days to 8 days)
Customer satisfaction raised from 87% to 96%; Sev 0–1 incidents cut from 8 to 3 a year
The self-tuning engine became the core product differentiator
Patent Application
Method, System and Device for Predicting Stock Performance and Building an Alert Model for Such Estimation
US 2022/0084119 A1 · Filed November 17, 2021
Describes a system that analyzes continuous time-series market data to identify and validate bullish "W" and bearish "M" patterns using precise shape and timing rules. It filters noise with multi-stage anomaly detection, forecasts reversal zones and entry and exit levels, and feeds validated patterns into adaptive models that update in near real time and trigger customized alerts.
WHAT I LEARNED :
Building the product and the organization at the same time taught me that engineering quality and team health can't be separated.
Introducing AI-assisted development also surfaced a real fear among engineers of being replaced. Adoption depended less on the tools than on clear expectations, accountability, and a culture that treats AI as a multiplier, not a threat. The biggest gains came from pairing new capabilities with human judgment: every change still gets human review.
TECHNOLOGIES :
C# · .NET · SQL Server · AWS (EKS) · In-memory caching · Parallel processing · Event-driven · Real-time systems · Adaptive algorithms
VIVIMETRIX PLATFORM
Architect (2021–2023), then CTO & Chief Architect (2024–present)
Real-time fintech SaaS platform with a self-tuning pattern recognition engine.
THE CHALLENGE :
Build a real-time trading platform and the engineering organization behind it, starting from nothing, in a crowded fintech market.
The platform had to read fast-moving market data across ~10,000 stocks every trading minute, reliably recognize complex chart patterns, alert subscribers in under a second, and stay dependable as both the customer base and the team grew.
APPROACH & KEY DECISIONS :
I treated the core problem as perception under noise, not just data processing. The engine had to "see" patterns in raw price data the way an experienced trader does, ignore the noise, and get better over time.
Key decisions:
Designed the pattern recognition engine from first principles instead of relying on existing libraries
Used geometric analysis (triangles and slope relationships) to separate real patterns from noise
Made the engine self-tuning: every day it scored every pattern, alerted and abandoned, against actual outcomes and adjusted its own parameters for the next session
Prioritized performance through in-memory design and parallel processing
Built the engineering team, processes, and culture alongside the product, not after it
SOLUTION & IMPLEMENTATION :
I architected the full platform and led its build while growing the engineering team. That included:
A pattern recognition engine that processes 10,000 charts in under 10 seconds
Real-time data pipelines and alerting with a sub-second delivery SLA
Cloud-native architecture on AWS (EKS) with strong observability
Multi-tenant isolation with row-level security and encryption, redundancy, and disaster recovery
PCI-compliant payments through an event-driven Stripe integration
Team structure, hiring, sprint process, CI/CD, and code quality practices
Through repeated optimization and caching, per-chart analysis went from multiple seconds to 5 ms in production.
RESULTS & IMPACT :
99.99% platform uptime
Pattern detection at 5 ms per chart, with alerts delivered in under 1 second
Architecture that carried 100% of revenue through the first $1M+ and several thousand users
78% average annual revenue growth since 2022; new pattern features lifted subscription revenue ~50% at launch
Engineering grown from 5 to 15 under four managers, with change lead time cut ~40% (14 days to 8 days)
Customer satisfaction raised from 87% to 96%; Sev 0–1 incidents cut from 8 to 3 a year
The self-tuning engine became the core product differentiator
Patent Application
Method, System and Device for Predicting Stock Performance and Building an Alert Model for Such Estimation
US 2022/0084119 A1 · Filed November 17, 2021
Describes a system that analyzes continuous time-series market data to identify and validate bullish "W" and bearish "M" patterns using precise shape and timing rules. It filters noise with multi-stage anomaly detection, forecasts reversal zones and entry and exit levels, and feeds validated patterns into adaptive models that update in near real time and trigger customized alerts.
WHAT I LEARNED :
Building the product and the organization at the same time taught me that engineering quality and team health can't be separated.
Introducing AI-assisted development also surfaced a real fear among engineers of being replaced. Adoption depended less on the tools than on clear expectations, accountability, and a culture that treats AI as a multiplier, not a threat. The biggest gains came from pairing new capabilities with human judgment: every change still gets human review.
TECHNOLOGIES :
C# · .NET · SQL Server · AWS (EKS) · In-memory caching · Parallel processing · Event-driven · Real-time systems · Adaptive algorithms