Technology Operations & Service Delivery·Engineering Leadership·AI Transformation
18+ years running enterprise platforms and global delivery teams at Bayer, Cognizant, JP Morgan Chase and CVS Health — and a hands-on builder of production AI voice platforms.
Hi, I'm Nana's assistant. Ask about his experience, leadership, AI work, or fit for your role, or pick a question above.
At Bayer I led global 24×7 operations for FieldView, a $750M+ digital farming platform in 23 countries: mean time to restore dropped from 96 hours to 90 minutes, SLA compliance rose from 40% to 85%, and I helped steer 400+ engineers through Bayer's SAFe transformation. Before that I led a 104-person global delivery organization at Cognizant, and India technology integration through the Bear Stearns, Washington Mutual and Bank of New York transactions at JP Morgan Chase. Today I lead service delivery for 42+ enterprise platforms at CVS Health.
Since 2025 I've also built AI products end to end, as founder of KnoxCalls LLC: four AI platforms in production, for home services, restaurants, recruiting and fraud protection, with four more in development. Building them keeps my AI judgment grounded in what actually ships, and that's what I bring to AI transformation inside an enterprise.
Platforms, people and P&L at scale, across healthcare, agtech, retail and banking.
Designed, built and operated as sole founder and engineer. Each one taught me something about running AI in production that a vendor demo doesn't.
Outcome: an AI voice agent triages after-hours calls for HVAC, plumbing and electrical businesses, pages on-call technicians for emergencies until someone acknowledges, and logs routine leads to a CRM.
How: real-time emergency classification, an escalation loop over SMS and voice, and a full observability stack with SLO-based alerting, run at 98%+ SLA.
Outcome: handles up to 20 concurrent calls, recognizes repeat callers and routes orders to the kitchen with no staff involvement.
How: menu import from a photo in under 60 seconds (GPT-4o Vision), Square and Toast POS integration, and subscription-plus-usage billing.
Outcome: runs inbound and outbound screening calls, scores candidates against per-role rubrics, books interviews and syncs to the ATS.
How: embedding-based matching with a GPT-4o reranker, encrypted PII, do-not-call enforcement and Greenhouse integration; 353 backend tests green.
Outcome: analyzes live calls and texts for fraud patterns (urgency, gift-card requests, impersonation) and alerts family members the moment a threat appears.
How: streaming speech-to-text plus an LLM threat classifier, with a native Android app that intercepts live calls.
What: outbound calling campaigns where each tenant switches on the Skills and premium Add-ons it needs.
How: built on Dograh, the open-source voice-agent platform, so telephony, speech-to-text and voice providers stay swappable instead of locked to one vendor.
What: answers calls as the owner's stand-in, while a companion Builder Agent runs coaching calls with the owner to deepen the persona over time.
How: layered prompt composition, voiceprint recognition for returning callers, and vector search over the owner's knowledge.
What: pulls listings from ATS boards into one feed, scores fit against a CV with an LLM, and drafts tailored application materials.
What: structured cognitive-training exercises with progress tracking. Early stage.
Hearth is caregiver coordination for adult children of aging parents: an honest weekly picture of how a parent is doing, built from a commodity sensor kit in the home. It isn't a product yet. What's worth showing is how it's being built.
AI reasoning will become a commodity; accountability won't. So the loop that acts on or reports about a person stays deliberately simple (a calibrated estimator, declarative rules, an audit trace), and "why did it do that?" always has a short answer. Frontier models work at the edges: understanding speech, narrating traces, writing the weekly summary. It makes no clinical claims; every sentence traces back to what was sensed.
Each model has a job and written limits on what it may change.
One ledger holds every decision and its rationale. Assumptions are listed so they can be tested and killed, and every session ends with a clean handoff.
It's the same operating model I'd bring to AI adoption on an enterprise team: clear roles, written decisions, audit trails, and a human accountable for every change.
Hardware meets biology: a low-cost field instrument co-invented with a postdoctoral researcher.
Spider-web vibration research normally relies on lab-bound laser vibrometers costing tens of thousands of dollars. This instrument takes comparable measurements in the field at about 1/100th of the cost, from off-the-shelf parts in a novel configuration.
Operations leadership first, with enough hands-on AI and cloud depth to lead the teams doing the work.
Open to VP / Director roles in technology operations, service delivery and engineering, and to senior AI engineering roles. Also open to consulting and partnership conversations around my AI products.