Ryan Yavari

Software Engineer / AI Engineer

I build production AI systems, RAG pipelines, and agentic applications. Currently pursuing Georgia Tech's OMSCS (AI specialization) while shipping real products.

SJSU CS grad, 2025. Open to relocation.

Ryan Yavari

Featured Work

SJC Airport AI Concierge

Production RAG-powered customer service chatbot for San Jose Mineta International Airport, live at flysanjose.com.

10,000+travelers served
99.99%uptime over 6+ months
66%latency reduction
75→98%accuracy improvement
17%call volume reduction
25%cost reduction

A tool-augmented RAG pipeline built with LangChain, OpenAI embeddings, and Qdrant, grounding responses across 13,000+ chunked documents pulled from 667 pages of the SJC website.

A Redis-based semantic cache with hash-value change detection and automated refresh, cutting inference costs by 25% and improving end-to-end response latency by 66%.

A model selection evaluation suite using an LLM-as-Judge approach across 200+ benchmark Q&A pairs, improving answer accuracy from 75% to 98% and reducing response latency from 5s to under 3s. Deployed serverless on Google Cloud Functions and Firebase with retry logic and graceful failure handling.

View live at flysanjose.com— chatbot is in the bottom right corner


Personal Projects

PrepAgent

AI Interview Research Agent

Sept. 2026 — Nov. 2026

A full-stack AI interview-prep app. You enter a company, role, and optionally an interviewer and a job description, and a LangGraph pipeline researches the company and interviewer on the web in parallel, retrieves your resume and prior same-company briefings from pgvector, and synthesizes a four-section briefing with Claude. Every talking point must cite a source URL or a resume section, and the backend validates those citations and drops anything uncited, so unsupported claims are visible instead of hidden. Google OAuth handles sign-in, and an optional Google Calendar connection detects interview events and proposes a briefing for you to confirm. Quality is tracked with a RAGAS evaluation harness (faithfulness, answer relevancy, context precision) with committed baselines, and the service is hardened with timeouts and retries, rate limits, per-user cost caps, and prompt-injection defenses. Built with FastAPI, Next.js, Supabase Postgres, Docker on AWS and Vercel, and GitHub Actions CI.

LangGraphClaudepgvectorFastAPINext.jsRAGASGoogle OAuth

Optimal

AI Productivity Agent

June 2025 — Jan. 2026

A supervisor-orchestrated multi-agent system with structured agent handoffs and human-in-the-loop approval gates, so agents propose actions for user review before anything gets written. Built offline evaluation benchmarks and regression suites to measure agent task completion and output quality, cutting prompt regression incidents by 40% across 3 model update cycles. Full-stack with React Native (Expo), Spring Boot, and a Supabase PostgreSQL schema with 17+ entities. Distributed via TestFlight for real-world testing.

React Native (Expo)Spring BootSupabasePostgreSQL

Quizzler

AI Quiz Generator

Mar. 2025, Updated Apr. 2026

A 5-stage RAG pipeline instrumented with LangSmith tracing across query expansion, embedding, reranking, and generation, capturing token usage, Cohere rescores, and per-span latency to surface retrieval quality and cost regressions. Features parallel upload processing to chunk and embed user documents into an ephemeral server-side vector store while concurrently generating structured MCQ quizzes, delivering grounded answers with glass-box citations from user-uploaded PDFs and text.

LangSmithRAGCohere RerankVector Store

Skills

Languages

PythonSQLTypeScriptJavaScriptJavaGolang (Go)CC++HTMLCSS

AI & Machine Learning

AI AgentsGenerative AILLMsTool CallingMCPRAGPrompt EngineeringAgentic SystemsWorkflow AutomationResponsible AIAnthropic APIOpenAI APIFunction CallingLangChainLangGraphLLM EvalRAGASHallucination DetectionObservabilityVector DatabasesData Preparation

Core Frameworks

FastAPIFlaskNode.jsExpressReactNext.jsSpring BootReact Native (Expo)

Databases & Cloud

PostgreSQLMySQLRedisSupabasepgvectorQdrantPineconeMongoDBGCPAWSFirebaseAzure

Tools

RESTful APIsOAuthGitDockerCI/CDPostmanLangSmithLlamaIndexVercel AILinuxKubernetesSystem Design