Chris Lee

Architected and developed a production-ready, full-stack automated content creation system that generates vertical video slideshows from real estate listings and publishes them to Instagram on a configurable schedule. The system processes listings from a REST API, captures screenshots using headless browser automation, generates MP4 videos with FFmpeg, and manages the complete Instagram publishing workflow via the Instagram Graph API.

CONTRIBUTIONS
• Integrated Puppeteer for headless browser automation to capture 1080x1920 vertical screenshots from dynamic web pages.
• Built an FFmpeg-based video slideshow generation pipeline with configurable slide duration and transitions.
• Implemented Cloudinary integration for temporary video storage during the Instagram upload workflow.
• Created an automated cleanup system to manage disk space and prevent storage overflow.

CONTRIBUTIONS
• Implemented full multi-domain tenant enforcement in a FastAPI back end and comprehensive routing/CORS/auth test suites, bringing feature completion to roughly 89% and 100% test coverage for the epic.
• Authored the domain cutover plan and automation to move prod traffic from demo.sapienceai.co to pilot.sapienceai.co (GCP URL maps, SSL, DNS, Cloud Build bucket swap), including interactive migration/verification scripts and quick-start/rollback playbooks.
• Designed and documented Vertex AI Data Store–based RAG pipeline with Dialogflow CX agents, Firestore metadata, and GCS ingestion, and provided end-to-end API/dataflow guidance for grounded chat responses.
• Conducted end-to-end A2A protocol review across the front end, back end, shared library, and membership agent, surfacing nginx timeout bugs, unused direct-connect path, SSE format gaps, and session/TTL consistency; produced remediation recommendations for low-latency agent streaming.
• Built a secure local developer experience: HTTPS nginx reverse proxy, Docker Compose orchestration, self-signed cert automation, and env templates for both services, enabling hot-reload workflow with a single docker-compose up.

CONTRIBUTIONS
• Built a FastAPI and PostgreSQL service that ingests Rent Comparability Study PDFs, runs a retrieval‑augmented extraction pipeline, and serves results via a REST API fronted by nginx; React/Vite UI consumes the API and Docker Compose runs the whole stack (back-end, front-end, DB, proxy).
• Supported authenticated, code‑gated user onboarding, cached per‑document PDF/embedding storage, and exported JSON/Excel outputs for HUD RCS fields.
• Designed the product to support ever-changing field requirements and incoming document types.

CONTRIBUTIONS
• Architected and shipped the Workflows v2 Python SDK, letting teams compose DAG-based tasks and configs, run them locally for dev or distribute across Celery/Redis via Jetstream for production-grade ML/document workflows (Python 3, Pydantic, msgpack, Docker).
• Led Jetstream evolution into the platform’s distributed task runner—hardening Celery configs, health checks, heartbeats, msgpack serialization, and Redis/RabbitMQ tuning to keep high-throughput microservices reliable at scale.
• Built the Indico platform FastAPI module set (auth, cookie signing, DB connectors, logging, testing harnesses) so new services inherit secure JWT flows, signed cookies, pooled database access, and uniform observability out of the box.
• Delivered the Workbench (“Gallery”) service: a FastAPI and PostgreSQL microservice for managing versioned field blueprints with tagging, search, bulk import/export, and audit trails—standardizing ML data schemas across apps.
• Drove cloud-native practices across services (Poetry/uv for deps, Docker Compose for local stacks, pytest suites with fixtures) to keep dozens of microservices production-ready and easily reproducible.

CONTRIBUTIONS
• Built HobbyBox, a mobile and web platform for trading-card collectors to discover shows, follow dealers, chat in event rooms, and share inventory via public profiles.
• Led a FastAPI back end (api-server) that serves cards, events, trades, search, notifications, and analytics, with auth, caching, and pagination baked in.
• Designed an S3/SQS-driven card-upload pipeline that hands off images to ML services (YOLOv9 detection, PaddleOCR extraction, verification) to auto-create PSA slab entries and keep status webhooks in sync.
• Shipped the React Native/Expo app (front-end/hobby-box-app) with NativeWind UI, OTA release flow, and CometChat-powered event group chats; paired it with a TanStack Router/Tailwind web experience (hb-web) for shareable profiles.
• Added supporting services and Lambdas for price comps, certification lookups, event sync, push notifications, and email campaigns to keep collectors engaged across channels.