Noa Sasson

Led full-stack builds across web, AI, and automation products, delivering onboarding systems, geo-driven analytics dashboards, internal tooling, multi-LLM pipelines, and multi-agent automation. • Built onboarding SDKs, templates, and documentation across healthtech, legal AI, and internal tooling, cutting time-to-first-success by 50% and driving >20% WAU conversion in 30 days • Designed prompt-tracing and diagnostic tooling for RAG pipelines, reducing LLM issue-resolution time by 70% and increasing engineering iteration velocity • Integrated OpenTelemetry and structured logging into LLM-facing APIs, improving production visibility into latency, failure modes, and token usage • Scaled developer communities from 0 to 2.4K+, running weekly technical sessions and surfacing 100+ actionable issues across early adopters • Authored technical content and workshops viewed by 25K+ developers, adopted as onboarding collateral by multiple clients • Expanded OSS contributor base from 3 to 50+ via contributor docs, sprint planning, and GitHub-first workflows while maintaining an >85% merge rate • Built multi-LLM orchestration pipelines for fintech and SaaS use cases, including financial assistants and multi-source document Q&A with function-calling across 3+ models • Developed Python and Django APIs supporting LLM-driven recommendation and analysis features, reducing NLP pipeline latency by 45% via token c • Deployed full-stack GenAI products on cloud inference platforms with token-level latency under 300ms and uptime above 99.95% • Developed legal-tech RAG systems indexing 10M+ documents with sub-1.8s response latency and 92% hallucination-suppression on benchmark prompts • Integrated LLM features into enterprise SaaS via secure JSON APIs, embedding RBAC, audit trails, and usage quotas to meet SOC2/GDPR requirements • Produced 15+ architecture deliverables (HLDs, integration maps) that shortened pre-sales and onboarding cycles by 40% • Benchmarked GCP-native vs hybrid infra for LLM workloads across TPU/GPU tiers with ±5% variance targets on latency and throughput • Built GenAI MVP stacks for early-stage teams, including streaming inference layers, retriever-selection models, and RAG pipelines tailored to time-to-market constraints • Built modular agent-orchestration layers with retriever-memory combinations and tool-chaining logic, supporting 12K+ evaluations with <2.3% output variance • Designed multi-agent workflows for content generation, triage, and analytics, improving throughput by 3.8× and stabilizing outputs across distributed pipelines • Implemented vector pipelines connected to CRM and analytics systems, reducing duplicate inference and lowering API costs by 31% • Built automation layers using n8n, Zapier, LLM agents, and vector databases, reclaiming 240+ staff hours monthly and influencing $620K+ in revenue pipeline • Created Streamlit and Retool control panels exposing real-time agent telemetry, cutting troubleshooting cycles from hours to minutes