Restaurant recommendation service

Designed and delivered a reference implementation that enriches restaurant listings with human-readable tags derived from reviews, photos, and restaurant websites/menus. Implemented a geospatial bootstrapping strategy using H3 hex indexing to expand data collection outward from a user's search region.

Role

Principal engineer

Timeframe

May to June 2025 (time-boxed engagement)

Outcomes

• Validated feasibility and key constraints early with a working demo.
• Surfaced API throttling and concurrency constraints for prompt/model tuning.
• Delivered a reference implementation plus a prioritized backlog for handoff.

Stack

• AWS Lambda, S3, SQS, PostgreSQL
• AWS Bedrock (Claude Sonnet)
• Node.js, Puppeteer, Docker, AWS SAM
• H3 hex indexing

Writeup

Built an automated enrichment pipeline that converts unstructured text and image signals into structured tags. Used headless Chrome automation to capture menu and website content, then ran LLM-assisted parsing for tag generation. Implemented H3-based crawl expansion to bootstrap coverage, plus SQS-triggered fan-out workflows for per-restaurant enrichment.