Success Stories
Strategy through implementation — here's what that looks like in practice.
Cycle time reduction after embedding AI across the SDLC.
Requirements, coding, QA automation, release ops.
Time from idea to working detection system.
AI-native SDLC for a sportstech platform
I worked with a regulated sportstech operator to embed AI into every step of their software development lifecycle. I shaped the AI strategy with leadership, then embedded with the engineering teams to implement it — from fine-tuning dev tools to building deployment guardrails.
Highlights
- Defined the AI adoption roadmap with the CTO, then built it with the teams.
- Unified AI workspace tuned to their domain-specific microservices.
- Spec-driven development orchestrated inside Linear keeps product, engineering, and QA aligned.
What I Built
- Cursor fine-tuning pipeline aligned to the company's code conventions and architecture.
- AI-assisted requirements templates that convert betting concepts into executable specs.
- Automated QA harness combining synthetic data generation with regression packs.
- Deployment guardrails so AI-authored code moves through review, testing, and release safely.
Results
- 50% faster delivery velocity across prioritized squads.
- QA automation covering mission-critical journeys.
- Consistent developer adoption because workflows stayed inside existing tools.
How It Happened
Discovery
Shadowed product, engineering, and compliance teams to map where AI could remove friction.
Build
Fine-tuned Cursor, authored prompt libraries, paired with engineers to validate outputs in their actual codebase.
Scale
Rolled out enablement playbooks, observability dashboards, and continuous tuning cadences.
Romance-scam detection — from concept to working prototype
An early-stage investor needed to validate a concept: could AI detect romance scams before money moves? I didn't just research it — I built a working prototype to prove (or disprove) the thesis in six weeks.
Highlights
- Built a prototype scoring engine that classifies offender patterns in near real-time.
- Ran rapid experiments using LLMs plus behavioral graphs to test detection approaches.
- Mapped the full attack surface — human, technical, and regulatory — in under six weeks.
What I Built
- Working prototype that ranks romance-scam risk from messaging patterns.
- Signal catalogue blending scraped intelligence, partner datasets, and user interviews.
- Threat model covering dating apps, messaging platforms, and payment rails.
- Technical feasibility assessment with clear next steps.
Results
- Prototype proved the core detection approach works.
- Clear differentiation narrative anchored in novel behavioral signals.
- Investor had enough evidence to proceed with staged build.
How It Happened
Research
Gathered qualitative interviews, dark-web intel, and incident data to understand the problem space.
Build
Built lightweight detection models, tested precision vs. false positives, iterated on the approach.
Validate
Demonstrated the prototype, documented findings, outlined the path forward.