TL;DR - Six years ago I was a data science intern writing risk models in R. Today I am an Associate Director leading AI/ML at a global ratings agency, running two SaaS products, and shipping agentic systems into production. This post maps the pivots - from insurance analytics to full-stack engineering to trading systems to entrepreneurship - and the one pattern that held across every transition: technical breadth compounds into leadership leverage.
The Arc in One Diagram
Before diving into each phase, here is the full timeline. Each pivot was triggered by the same signal: the work I was doing no longer demanded the full range of what I could build.
Career arc - six years in four pivots
Phase 1: Data Science Foundations (2019)
My career began at a financial services firm as a data science intern. The team was building insurance risk models - logistic regressions, gradient-boosted trees, and survival analysis against actuarial datasets. I wrote a lot of R, some Python, and spent evenings learning why a model that scored 0.94 AUC in training could collapse to random guessing in production.
The key insight from this phase was not about statistics. It was about feedback loops. A model takes days to train and weeks to validate. A UI change takes seconds to evaluate. I wanted tighter loops.
What I learned at a financial services firm
Phase 2: Full-Stack Engineering (2020–2022)
At a healthcare analytics firm, I joined as a software developer building the platform layer that made healthcare data accessible to sales and marketing teams. This was my first real exposure to React, PostgreSQL, AWS, and the discipline of shipping code that real users depended on daily.
// The kind of query I learned to write daily
const getHealthcareFacilities = async (filters: FilterSet) => {
const query = buildDynamicQuery(filters)
// paginated, indexed, cached at the API layer
return await db.query(query).paginate(50)
}The biggest technical lesson here was abstraction debt. Our early monolithic API became impossible to extend after 200+ endpoints. We split into domain services - provider data, claims, affiliations - and the velocity difference was immediate.
Key lesson - Monolith velocity is an illusion. The first 50 endpoints are fast. The next 150 will make you wish you had split at 20.
Phase 3: Financial Systems at Scale (2022–2024)
Joining a global investment bank as a post-trade systems developer was a leap into high-stakes engineering. Our systems processed millions in daily transactions. A single bug could mean a regulatory breach. This environment taught me that resilience is not a feature - it is the product.
I worked on a microservices architecture that handled trade confirmation, settlement, and regulatory reporting. The stack included Java, Kafka, Oracle DB, and internal event-driven frameworks. Every service had to tolerate downstream failures without cascading.
Resilience patterns from trading systems
The lesson that stuck: observability is the foundation of reliability. Without structured logs, distributed traces, and latency dashboards, you are debugging in the dark. I carry this into every project I touch today.
Phase 4: Entrepreneurship (2021–Present)
Parallel to my corporate career, I started building SaaS products.xThreads.app - an AI-powered Twitter content tool - gained 1,000+ users and taught me the full product lifecycle from idea to deployment to support.PostQueue.app followed, adding multi-platform scheduling with AI translation and optimization.
// The stack that shipped both products
const stack = {
frontend: "Next.js + Tailwind",
backend: "FastAPI + PostgreSQL",
ai: "OpenAI API + LangChain",
infra: "Docker → AWS ECS Fargate",
}
// One-person team. Zero ops team. Everything automated.Key lesson - Speed to market beats technical perfection every time. A working prototype in users' hands is worth more than a beautifully architected system that nobody has seen.
Phase 5: AI Leadership (2024–Present)
Today, as Associate Director, Applied AI/ML at a global ratings agency, I lead the strategy and implementation of AI systems that serve financial analysts and clients. The role is a synthesis of everything that came before: the data science foundation, the engineering discipline from BNP, the product instincts from my SaaS work.
I now focus on agentic AI - systems that plan, execute, and self-correct across multiple tools and data sources. These are not chatbots. They are autonomous problem-solvers that operate within enterprise guardrails.
Summary: What Six Years Taught Me
- Technical breadth compounds. Every pivot taught transferable skills that made the next role easier, not harder.
- Production is the real teacher. Notebooks and prototypes teach patterns. Production teaches discipline - monitoring, retries, degraded modes, incident response.
- Domain knowledge is a force multiplier. Understanding fintech, insurance, or healthcare makes you 10× more valuable than someone who only knows the framework.
- Ship. Then iterate. Every successful product I have built started as something embarrassingly simple that solved one real problem well.
- Leadership is about leverage. The most senior engineers I have worked with do not write the most code. They choose what gets built, how it gets built, and who builds it.
The next chapter is about agentic systems in production - bridging the gap between what AI can demo and what enterprises can trust. The combination of enterprise rigour and entrepreneurial speed has never felt more relevant.