Engineering Choices You Have to Defend

Episode Summary:In this episode of Engineering Choices You Have to Defend, host Nicola Onassis sits down with Roy Resh, VP of Engineering at Trax Retail, to explore a pivotal architectural decision that reshaped how large-scale computer vision systems are built and scaled in retail environments.At Trax, Roy and his team built a computer vision platform that analyzes shelf images captured in retail stores, identifying products, pricing, and point-of-sale materials to generate a digital representation of store shelves. This enables brands to measure execution, shelf share, and product availability in near real time. But as the platform scaled across enterprise clients, complexity began to compound rapidly.What started as a unified recognition pipeline evolved into a heavily customized system, with per-client logic for attributes like expiration dates, display detection, reporting formats, and KPI calculations. Each new customer introduced new requirements, leading to custom code per client, duplicated processing flows, and increasingly long onboarding cycles that stretched from weeks to months.Roy explains how the system eventually reached a breaking point: onboarding delays of 30–60 days, rising operational overhead, and microservices becoming entangled with client-specific logic. In some cases, the platform even processed the same image multiple times to satisfy different customer requirements, driving up cost and complexity.The team made a strategic decision to move away from custom implementations and toward a configurable, JSON-driven workflow architecture. Built on event-driven microservices, queues, and coordination barriers, this new system allowed engineering teams to define and version entire processing flows through configuration rather than code.This shift enabled safer deployments, faster experimentation, and gradual rollouts per client—without affecting the entire platform. It also introduced a standardized KPI layer, reducing the need for bespoke reporting logic across customers.Roy also discusses the importance of human-in-the-loop validation in production AI systems. In a constantly evolving retail environment, human annotators help generate training data, validate model outputs, and maintain accuracy for high-stakes enterprise use cases where precision is critical.For engineering leaders, this episode highlights a key lesson: when every customer forces new code paths, you’re not scaling a product—you’re scaling complexity.Key Takeaways:Over-customization is a clear signal of architectural scaling limitsLong onboarding cycles often reveal hidden system fragmentationConfigurable workflows reduce dependency on per-client code changesEvent-driven, JSON-based orchestration improves flexibility and deployment safetyGradual migration strategies reduce risk in enterprise system rewritesStandardizing KPI logic is as important as standardizing AI pipelinesHuman-in-the-loop systems remain essential in dynamic real-world AI environmentsScalable platforms reduce variability instead of multiplying itConnect with Roy Resh:LinkedIn: Roy Resh: linkedin.com/in/roy-reshListen Now & Subscribe:Apple Podcasts, Spotify, Amazon Music, or wherever you get your podcasts."Engineering Choices You Have to Defend explores the real technical decisions behind regulated software, compliance, and AI integration, helping leaders build secure, auditable, and user-friendly systems."

Episode Summary:In this episode of Engineering Choices You Have to Defend, host Nicola Onassis sits down with Keith Deming, an engineering leader with experience at Postmates, Uber, and PRISM Skylabs, to explore a pivotal architectural decision that transformed how computer vision systems scale in the real world.At PRISM Skylabs, Keith and his team built a platform that turned retail surveillance cameras into powerful analytics tools, tracking foot traffic, customer journeys, and in-store engagement. The system worked exceptionally well… until customers wanted it everywhere. What started as a four-camera deployment quickly became a 200-camera scaling challenge, exposing the limits of server-based infrastructure.Keith shares how the team faced mounting constraints, hardware costs, power consumption, cooling limitations, and physical space, and realized that simply scaling servers wasn’t viable. Instead, they made a bold shift: moving compute from centralized servers directly onto the cameras themselves.The conversation dives into how a Raspberry Pi prototype proved edge computing was feasible, why rewriting performance-critical systems from Python to C++ became necessary, and how eliminating video decoding overhead unlocked real-time processing. More importantly, this architectural shift didn’t just solve a technical problem, it removed friction from the buying process, making it easier for customers to adopt and scale the product incrementally.Keith also reflects on how modern advancements in edge AI and distributed computing are reshaping system design today, and why many teams still underestimate the true cost of centralized infrastructure.For engineering leaders, this episode highlights a critical lesson: scaling isn’t always about adding more resources—it’s about rethinking where computation happens.Key Takeaways:Centralized infrastructure can become the biggest bottleneck to scaleEdge computing eliminates hardware, power, and space constraintsMoving the compute closer to the data reduces latency and processing overheadPrototyping with simple tools (like Raspberry Pi) can unlock major breakthroughsRewriting for performance (Python → C++) is often necessary at scaleRemoving infrastructure friction accelerates customer adoptionThe best architectures reduce reasons for customers to say “no”Distributed and edge-based systems are becoming the future of AI deploymentConnect with Keith Deming:LinkedIn: https://www.linkedin.com/in/keith-demingListen Now & Subscribe:Apple Podcasts, Spotify, Amazon Music, or wherever you get your podcasts."Engineering Choices You Have to Defend explores the real technical decisions behind regulated software, compliance, and AI integration, helping leaders build secure, auditable, and user-friendly systems."

Episode Summary:In this episode of Engineering Choices You Have to Defend, host Nicola Onassis sits down with Sean Graham, VP of Engineering at Idelic, to unpack a critical shift in how engineering teams approach delivery in high-stakes environments.At Idelic, where software directly impacts fleet safety, compliance, and insurance risk, reliability isn’t optional. Sean shares how their team moved away from traditional two-week sprint cycles after realizing that large batch releases were quietly increasing risk. While velocity appeared healthy on the surface, debugging became guesswork, QA was overwhelmed, and every deployment felt like a high-stakes event.Instead of optimizing Scrum, the team reframed the problem entirely, focusing on reducing batch size and risk. By shifting to a continuous, small-batch delivery model, they dramatically improved traceability, simplified debugging, and restored trust in their system. Lead time dropped from 25 days to just 4, while releases became routine instead of stressful.The conversation also explores how infrastructure, like per-ticket test environments and fast pipelines, enabled this transformation, and why discipline became the most important skill once sprint boundaries disappeared.As AI accelerates code generation, Sean emphasizes that structured delivery systems are more critical than ever. Without them, faster output simply compounds risk. Teams that pair AI with disciplined, low-risk delivery models will scale safely, while others risk creating faster chaos.For engineering leaders, this episode is a powerful reminder: speed isn’t about working harder, it’s about reducing risk and improving feedback loops.Key Takeaways:Large batch releases increase risk and reduce system reliabilityDebugging becomes exponentially harder when too many changes ship togetherContinuous, small-batch delivery improves traceability and confidenceLead time can drop significantly with continuous validation (25 → 4 days)Psychological safety and trust are critical for high-performing teamsStrong infrastructure is required to support fast, safe deliveryAI increases output—but without discipline, it also increases riskConnect with Sean Graham:LinkedIn: https://www.linkedin.com/in/sean-graham-675a054Website: https://profed.laroche.eduListen Now & Subscribe:Apple Podcasts, Spotify, Amazon Music, or wherever you get your podcasts."Engineering Choices You Have to Defend explores the real technical decisions behind regulated software, compliance, and AI integration, helping leaders build secure, auditable, and user-friendly systems."

Episode Summary:In this episode of Engineering Choices You Have to Defend, host Nicola Onassis sits down with Kevin DiGilio, President of KMD Technology. Kevin explains how compliance frameworks like ITAR, NIST, and DFARS don’t just guide documentation; they dictate core system architecture.When regulations evolved, KMD faced a choice: layer compliance on top of existing software or refactor the entire platform. They chose the latter, embedding user classification, role-based permissions, encryption, and access control throughout the stack. Kevin shares the trade-offs between usability and security, explaining how granular permissions and clear data classification maintain operational efficiency while staying fully compliant.The conversation also explores AI in regulated manufacturing environments. Kevin highlights how AI systems must inherit compliance rules, log every decision, and enforce strict data boundaries. Improper access or hallucinations aren’t minor—they can be catastrophic.For founders and engineering leaders, Kevin emphasizes that compliance should shape architecture from the start. Delaying integration almost guarantees costly rewrites, while proactive planning ensures systems that are secure, auditable, and operationally smooth.Key Takeaways:Compliance must be embedded into core architectureRole-based permissions balance usability and securityEncryption and access control are essential at every layerAI must respect regulatory boundaries with full logging and citation trackingDelaying compliance leads to costly refactorsConnect with Kevin DiGilio:LinkedIn: https://www.linkedin.com/in/kevindigilioCompany: https://kmdtechnology.com/Listen Now & Subscribe:Apple Podcasts, Spotify, Amazon Music, or wherever you get your podcasts."Engineering Choices You Have to Defend explores the real technical decisions behind regulated software, compliance, and AI integration, helping leaders build secure, auditable, and user-friendly systems."