Adobe — Software Engineer
New York, NY
Adobe — Software Engineer
- Engineered an agent tracing pipeline capturing 100K+ monthly traces via LangFuse, then fine-tuned agent LLMs on the labeled data, boosting task success rate from 25% to 70% and cutting prompt tuning from weeks to hours.
- Developed internal and public-facing APIs in Java and FastAPI as the backend systems powering the tracing, evaluation, and recommendation pipelines.
- Deployed and operated cloud-native infrastructure using Azure Functions, Blob Storage, and AWS Bedrock to run the production data and AI workloads behind those pipelines.
- Built an agent evaluation system with guardrails against unstructured outputs and silent failures, increasing result relevance by 45% and cutting undetected failures by 75%.
- Architected a reusable agent orchestration harness in CrewAI and LangGraph, standardizing cost tracking and observability across teams and cutting agent setup time from days to minutes.
- Built an agentic search service and Adobe's first context-aware recommendation app by having agents select search facets from document and parent-app context, driving adoption across 380 customer accounts.
- Built an automated security alerting system that detects and strips prompt injection and malicious content before it reaches the LLM, cutting triage time 60% and catching 20K+ risks monthly.
- Built AI-enhanced UI features in TypeScript and ReactJS, including recent-search components and suggested-content panels for agentic experiences.