• Led QA team in coordination with Behavior Scientist, UX designer, advisors and CEO to bring COPPA compliant vision for AI-assisted coaching to life.
• Built multi-tenant auth layer supporting SMS, SSO, email, OTP, & backend batch team imports; with CORS / CSP rules, & strict RBAC permissions on all content, all isolated per org subdomain.
• Designed centralized entitlements system connecting Stripe payments to each step of a 12-week course.
• Integrated Azure Cognitive Services for live voice assessment and lesson facilitation; OpenAI with server-side content orchestration.
• Observability & product analytics: Instrumented client and server with PostHog so feature usage and funnels are measurable in production.
• Supported iOS, Android and web with ReactJS, Material-UI, ReCharts & Flutter / DART to handle Deep Links, SafeArea, Push Notifications and OAuth login from native social app sessions.
• Automated end-to-end regression coverage with Playwright, producing videos with close captioning to publish as part of system health page.

• Designed AI orchestration pipelines that classify intent, gather context, trigger synchronous and asynchronous LLM/retrieval calls, and stream results over a single persistent connection via event-based messaging.
• Built secure and seamless guest-to-user conversion: anonymous chat history and uploads persist into the registered account when a visitor signs up.
• Created tmt-mcp, an open-source MCP orchestrator that drives coding and DevOps through GitHub Issues, raising release frequency 2.5× in its first week with zero bugs introduced.
• Delivered a CMS with QA and annotation tools to review application flows, empowering non-technical team members to QA, annotate LLM responses, and monitor token usage per chat intents.
• Supported team vibe-coders to turn prototypes into production-ready features.

• Maintained Django APIs and MySQL db powering ~150 write requests / second (+ EventBridge triggers after insert) - when hundreds of offline apps resync at once; with worker pool via Lambda auto-scaling. Images uploaded to S3 synchronously with 3 staggered retries.
• Built a server-side pipeline that turns campaign form templates into on-device redaction descriptors (SIFT keypoints + field geometry), so canvassers’ phones can redact PII and verify fields are filled out.
• Optimized SQL queries and aggregation endpoints for analytics dashboard, doubling data output while maintaining speed and adding search filters.
• Centralized logging through Firebase and wrote BigQuery monitoring to compare with SQL data for integrity and usage.

• Developed real-time canvasser tracking interface with MapBox and ReactJS, synchronizing map pins/popups and sidebar elements for seamless navigation. Programmed filter operations to recenter map boundaries with dynamic debounce based on filter in use. Built map to render any combination of layers including points for canvassers, lines for routes, and polygons for coverage area.
• Enhanced SASS implementation with mixins and reusable style patterns. Created modular components styled from a single classname styling, as well as composable utility classes to normalize spacing and typography while maintaining DRY principles and reducing CSS bloat.
• Restructured Redux implementation to flattening deeply nested data structures, enabling targeted updates and reducing component render cycles across decoupled components.
• Wrote responsive Data Grid component to render tables using Flexbox with hooks to customize any header or body cell's styles and click / hover events. Supported both client or server side sorting, pagination, and toggling visibility of any column. Wrote NodeJS script to preprocess API response schemas to ease client side table and list displays with custom property names and easier data aggregations baesd on data types (numeric, dates, enums).

• Built offline-first iOS and Android apps for 1200 canvassers nationwide to run Site and Door to Door campaigns following scripted intake forms.
• Shipped a C++ OpenCV native module (iOS + Android) that registers camera photos to the form, redacts PII fields, validates required signatures via ink detection based on pre-trained differentiation threshold - all before upload, entirely clientside, offline.
• Built components for background audio recording, live voter lookup registration, QR / barcode / document scanning, signatures and more.
• Battery / Memory Optimization: tuned background location tracking around device speed / distance, improved Redux state management to reduce render cycles, merged dependencies to reduce APK size by 67%

• Built API and scripts to generate embeddings of knowledge base - txt, pdf, ppt, docx - common questions, and user prompts.
• Built Docker orchestration of 6 containers to run in dev, cloud and onprem environments including behind airgapped SCIF networks.

AI Annotation Interface & APIs

• Created custom Node-Red modules for configuring cameras and communication with on-prem tools and cloud APIs.
• Implemented API for motion detection using Structural Similarity, to expose 19 OpenCV filter options to training interface, to manage photo training data and find tagging anomalies.
• Built entire front-end annotation tool with zooming, version controls, infinite class nesting.

Workstation Guidance

• Developed workstation guidance system for factories to configure assembly instructions with QA and computer vision assistance, dockerized to deploy across a local network of factory computers.
• Built bi-directional persistent secure WebSocket connection for constant communication with Node-Red service and FlexibleVision as a whole.

• Scripted OpenAI agent to recommend database schema for any idea, return both reasoning and a structured JSON output.
• Wrote utilities to parse Types spreadsheet and auto-generate React forms, Django app, fake data generator, Cypress.io tests, Swagger documentation and centralized data definitions to follow.
• Wrote tools to parse RBAC permissions matrix spreadsheet and produce strict TypeScript definitions and access control functions and Cypress.io tests.
• Wrote gcloud build scripts to stand up complete GCP environments based on shared .env variables.