I build AI products that people actually use.

The Problem

Healthcare clinics lose 30% of inbound calls due to staff shortages and after-hours gaps. Patients get frustrated, clinics lose revenue, and competitors with better availability win.

My Approach

• 0-to-1 product discovery: Interviewed 20+ clinic owners to validate pain points and willingness to pay
• Technical architecture: Designed LLM orchestration pipeline (ASR → intent classification → LLM routing → TTS) with sub-3s latency
• Build-vs-buy decisions: Evaluated 8+ AI providers across cost, HIPAA compliance, and latency. Chose hybrid approach (3rd-party voice + custom LLM layer)
• GTM from scratch: Built live voice demo, tiered pricing ($499/mo starting tier), and inbound funnel with no marketing budget
• Enterprise pilot program: Designed onboarding flow, SLA framework, and feedback loops for first 10 customers

Impact

Sole founder — owned product, engineering decisions, pricing, and first customer conversations. This is my biggest differentiator: I didn't just PM an AI product, I built one.

The Problem

TD Bank faced increasing insider fraud risk across 100K+ employees. Existing detection was manual, reactive, and failed to catch sophisticated patterns. The ask: build a platform that scales across multiple business lines without creating alert fatigue.

My Approach

• Cross-functional alignment: Ran workshops with 12+ senior stakeholders across fraud, compliance, legal, and engineering to define success metrics and non-negotiables
• Build-vs-buy evaluation: Led vendor assessment of 5 AI/ML platforms. Chose hybrid approach: 3rd-party ML engine + custom business logic layer
• Roadmap prioritization: Sequenced rollout across 5 product areas based on fraud impact vs. implementation complexity. Delivered MVP in 4 months vs. 12-month original estimate
• Performance tuning: Worked with data science team to optimize false positive rate from 40% to 8% through iterative model retraining and rule refinement

Impact

The Problem

TD's digital banking platform had generic, one-size-fits-all content. Users weren't discovering relevant products, leading to low engagement and missed cross-sell opportunities. The business wanted personalization but had no ML infrastructure or experimentation culture.

My Approach

• Experimentation framework: Built org-wide A/B testing standards — test design, sample size calculation, evaluation metrics. Ran 30+ experiments in first year
• Funnel optimization: Identified drop-off points across 5-step onboarding flow. Redesigned CTAs, copy, and placement based on user segments — improved activation by 12%
• ML model collaboration: Partnered with data science to define features, labels, and success metrics for recommendation engine. Iterated on 4 model versions to optimize click-through vs. conversion
• 0-to-1 innovation pipeline: Built business cases for new AI initiatives, presented to VPs, secured budget for projects with no precedent

Impact

Created experimentation playbook that became org-wide standard. Framework now used by 8+ product teams across the bank.

The Problem

The organization lacked a repeatable experimentation process. Product teams made decisions based on intuition rather than data, leading to inconsistent results and slow iteration cycles.

My Approach

• Framework design: Built comprehensive A/B testing standards covering test design, sample size calculation, statistical significance thresholds, and evaluation metrics
• Knowledge transfer: Trained product teams across the org through workshops and documentation — made experimentation accessible to non-technical PMs
• Embedded rituals: Created weekly experiment review cadences and shared learnings across teams to build institutional knowledge
• Tool selection: Evaluated and implemented experimentation platforms that integrated with existing tech stack

Impact

Framework became org-wide standard, reducing decision cycles and improving growth velocity across the bank's digital products.

The Problem

A critical 5-step onboarding funnel showed significant drop-off rates between steps. Users were abandoning before reaching activation, impacting acquisition efficiency and customer lifetime value.

My Approach

• Segmentation analysis: Analyzed user behavior across demographics, entry points, and device types to identify high-value segments with worst drop-off
• UX testing: Ran moderated user tests and session recordings to uncover friction points — discovered confusing copy and poor mobile experience
• Hypothesis-driven redesign: Redesigned CTAs, simplified copy, and optimized placement based on user eye-tracking patterns
• A/B validation: Tested changes incrementally across segments to isolate impact and avoid regression

Impact

12% activation improvement translated to measurable user growth and provided a playbook for future funnel optimization work across other product surfaces.

The Problem

Needed to balance performance, cost, and compliance for AI infrastructure. Wrong choice would impact speed-to-market, scalability, and total cost of ownership. For JustReva: HIPAA compliance was non-negotiable. For TD: enterprise security and auditability were table stakes.

My Approach

• Comprehensive vendor analysis: Evaluated 8+ AI providers for JustReva (ASR, TTS, LLM) and 5+ ML platforms for TD fraud detection — scored across latency, cost per transaction, compliance certifications, and API extensibility
• Hybrid architecture decisions: Chose 3rd-party voice engines + custom LLM layer for JustReva to balance speed and differentiation. Chose 3rd-party ML engine + custom business logic for TD to minimize ops overhead
• Total cost modeling: Built multi-year TCO models factoring licensing, compute, and engineering maintenance — presented trade-offs to stakeholders
• Risk mitigation: Designed vendor lock-in escape hatches through abstraction layers and fallback options

Impact

Architecture aligned with long-term platform strategy, reduced TCO, and enabled faster iteration through smart vendor partnerships while maintaining control over core differentiation.

The Problem

Conflicting priorities across fraud, legal, compliance, and engineering threatened delivery timelines for the fraud detection platform. Each function had legitimate but competing requirements, and lack of alignment risked scope creep and delays.

My Approach

• Stakeholder workshops: Ran structured alignment sessions with 12+ senior leaders — surfaced hidden dependencies and non-negotiable constraints early
• Shared success metrics: Defined common KPIs that all functions could rally around (e.g., fraud coverage + false positive rate + time-to-alert)
• Transparent trade-offs: Built decision frameworks that made trade-offs visible — when legal needed audit trails, showed impact on performance and presented options
• Iterative planning artifacts: Created living roadmaps and risk registers accessible to all stakeholders — reduced "surprise" objections late in delivery

Impact

Delivered MVP 8 months ahead of original estimate with full stakeholder buy-in. Alignment framework became template for future cross-functional platform work.

The Problem

Voice AI for healthcare was a new problem space with little historical data. Needed to validate willingness-to-pay, critical use cases, and product-market fit before investing in technical build.

My Approach

• Deep user interviews: Conducted 20+ in-depth conversations with clinic owners and office managers — uncovered that after-hours coverage and staff shortage were bigger pain points than call efficiency
• Quantitative validation: Asked concrete pricing questions ("Would you pay $X/month?") rather than hypotheticals — validated $499/mo starting tier was acceptable for 60%+ of prospects
• Use case mapping: Translated qualitative findings into specific product requirements — prioritized appointment booking and basic triage over complex medical questions
• Competitive positioning: Identified gaps in existing solutions (generic chatbots lacked voice, medical answering services lacked AI) to carve differentiation

Impact

Validated problem-solution fit and willingness-to-pay before code was written. Discovery findings directly shaped technical architecture and pricing tiers, reducing build risk.

The Problem

As a solo founder with limited capital, needed to prove traction without marketing budget. Time-to-market was critical to validate the business model.

My Approach

• Interactive product demo: Built live voice demo that prospects could call immediately — removed friction of "book a demo" and let the product sell itself
• Transparent pricing: Published tiered SaaS pricing ($499/mo starter) upfront to qualify leads and signal legitimacy — no "contact sales" games
• Content-led inbound: Wrote targeted content addressing clinic owner pain points — distributed through healthcare forums and LinkedIn to build early evangelists
• Pilot-to-paid conversion: Designed frictionless pilot onboarding with clear success criteria — made conversion decision easy with usage data

Impact

First paying customers were acquired with zero paid acquisition spend. GTM strategy proved product-market fit and created repeatable playbook for scaling.