Ditto · Member of Technical Staff (Software Engineer) · Present

Production AI for Matchmaking

Ditto

Role

Member of Technical Staff (Software Engineer)

Timeline

May 2026 - Present

Location

San Francisco, CA

Stack

TypeScriptRedisBullMQBraintrustLLM Evals +1

Figures below describe work during May 2026 - Present; they are self-reported career results, not live product statistics.

Overview

AI Matchmaking Over iMessage

At Ditto, I build production systems for a consumer dating product: reliable message delivery, chatbot evaluation, and location-based matching.

Delivery

Reliable Multi-channel Delivery

Built and own SendIntent, the delivery layer for iMessage, SMS, and email. It preserves message ordering across 100+ phone lines and 3 providers during bursts of 400+ messages per minute, using Redis leases, epoch fencing, and durable per-line BullMQ queues.

Quality

Chatbot Quality and Latency

Built a release-gating harness with 6 Braintrust scorers: 3 deterministic and 3 LLM judges. Identified the AI stage as 85–90% of response latency and shipped skill pre-activation and no-reasoning fast paths.

Matching

Location-based Matching

Shipped the location picking and saving flow, H3 geo-indexed intersection matching, and an LLM venue-selection agent. The system was adopted across Summer 2026 matches to select real-world date spots.