Eric Kennedy

Senior Software Engineer | intuit.com

• Virtual Expert Platform: Built AI-powered marketplace, recommendation, and real-time workforce-optimization systems powering products including TurboTax Live and QuickBooks Live, connecting 4.2M customers with 26K experts at tax-season peak.
• AI matching: Scaled AI-driven matching from 0.2% to 36% of contact-center traffic across 1M+ annual interactions, delivering $11.4M in 2025 business impact by leading development of a platform using transcript analysis, embeddings, and agentic workflows. Patent filed.
• Automated routing: Delivered $14M in 2026 business impact by re-architecting AI matching into a fully automated, real-time routing system. Built a Kafka stream processor handling 3K+ messages per second that used historical transcripts to build an OpenSearch vector store, then matched incoming customer data against those interaction embeddings to generate personalized expert assignments at customer contact.
• Agent gateway: Eliminated thousands of hours of manual work for 90+ weekly active users by leading development of a hosted agent gateway for 20+ domain-specific AI agents across contact-center operations. Built the shared runtime, access controls, tool integrations, collaborative sessions, and interfaces for web, Slack, Claude Code, and service APIs.
• Transcript intelligence: Built two LLM systems over customer conversations. A Bring Your Own Prompt platform ran customizable analysis across thousands of transcripts for tax policy compliance detection with 95.5% precision, QA at scale, and expert skill gap analysis. A separate pipeline built an intent taxonomy for labeling and storing customer intents in OpenSearch for use across multiple models. It used a Faiss/HNSW semantic cache to reuse known taxonomy labels or surface new candidates.
• Observability: Spearheaded obs-as-code (Terraform/IaC for observability) to move alert delivery from a quarterly observability sprint into the regular development cycle. Colocated alert definitions with application code, increasing output from 10 alerts per quarter to 5 per week and eliminating the need for engineers to learn the Splunk UI.
• Reliability: Resolved a two-week production incident causing 2 overnight on-call pages per week by diagnosing and fixing a memory leak. Used heap dumps, thread dumps, and Prometheus to trace a cascade from heap OOM through consumer-thread exit and blocking queue deadlock to pod failure. Implemented graceful shutdown and Kubernetes SIGTERM rollback handling to prevent in-flight data corruption.

Founder | animepics.xyz

• Product: Built and launched a paid consumer GenAI product that turned customer photos into personalized anime portrait galleries. Owned the end-to-end experience from package selection and Stripe checkout through validated uploads, async GPU fulfillment, and email delivery.
• ML pipeline: Built a GPU-backed pipeline that fine-tuned a text-to-image diffusion model on each customer's uploaded photos to generate recognizable portraits across multiple anime styles. Customized training with dynamic epochs and regularization, then used PyTorch to merge each personalized model with selected style models before prompt-based generation, high-resolution correction, and upscaling.
• Frontend: Built a mobile-first purchase and upload flow with Next.js, TypeScript, Tailwind, Zustand, and React Query. Added client-side face detection and smart cropping to reject invalid photos and generate model-ready 512×512 inputs before training.
• Backend: Built a Python order and fulfillment system with Starlite and SQLAlchemy. Persisted order state in Postgres, stored photos in S3, and used Kafka for durable handoffs to GPU workers and back to the web service for gallery persistence and email delivery.
• Payments and auth: Built guest and authenticated checkout with Stripe, Google and Apple OAuth, and passwordless email. Customers could purchase before creating an account, then return securely to upload photos or view completed galleries.

Staff Software Engineer | ethoslife.com

• Early engineer: Helped scale Ethos from stealth to $159.8M in 2023 revenue and 100+ engineers after joining pre-launch as the company's 3rd engineer. Built foundational frontend and backend systems that became core company infrastructure.
• Technical leadership: Reached Ethos's highest IC level and mentored engineers through transitions into engineering management. Served as the final escalation point for major production incidents across the platform.
• Platform scaling: Supported Ethos's growth from one core engineering group into a multi-team organization by migrating production workloads from ECS to EKS with 100% uptime. Standardized deployment, service ownership, and observability across teams with Kubernetes and Datadog.
• Deploy automation: Reduced deploy time from about 5 hours to 5 minutes by building a GitOps pipeline with GitHub Actions, ECR, a Go deployment service, and Argo CD. Added a React admin UI that let operators map commits to releases.
• Service architecture: Mapped domain and data boundaries with ERDs to help decompose the core monolith and enable independent team ownership. Helped establish a microservice-first model as the platform grew to 20+ Kubernetes services.
• Database migration: Improved data modeling and feature delivery by migrating MongoDB to Postgres. Enforced schemas at write time and used ACID transactions to preserve integrity across financial workflows.
• Event platform: Led the architecture and built a shared event platform adopted by 5–10 services across 10–20 topics, handling 50K+ events per day. Used a transactional outbox with Postgres CDC to commit state changes and events together. Added Kafka, Avro schemas, and an HTTP dispatcher so teams gained replayable, versioned events without owning broker complexity.
• Owned additional platform and data-system work across database operations, application performance, data modeling, and query optimization:
  • Database operations: Owned production Postgres reliability, recovery, performance, and cost using Multi-AZ RDS, point-in-time recovery, pganalyze, pg_cron, reserved instances, and reusable Terraform modules.
  • Frontend performance: Improved page-load speed and made frontend systems easier to extend by adding application and Nginx asset caching, reducing redirects, and standardizing CORS and cookie handling across subdomains.
  • Backend performance: Improved API response times with reusable Redis middleware caching and prevented stale assets by building a Kubernetes operator that invalidated the CDN when images changed.
  • Data modeling: Strengthened referential integrity by replacing polymorphic relationships with exclusive-arc constraints that ensured each record referenced one valid parent. Shipped the schema change with zero downtime.
  • Query optimization: Reduced a payments query from 4 hours to 5 seconds by analyzing its execution plan and restructuring the SQL.