Active
RAG-Based Knowledge Retrieval System
Scalable Retrieval-Augmented Generation system with 100K+ embeddings using LangChain, FastAPI, and Weaviate. Optimized retrieval accuracy with chunking strategies and context window management.
Hey there!
If it's not in prod, it doesn't exist.
Latency is the silent killer of user trust.
RAG is easy. Good RAG is hard.
Context engineering > prompt engineering.
Agents are just while loops with ambition.
Trace everything. Assume nothing.
I'm Ziyan, building AI systems that ship to production
GenAI Engineer specializing in LLM systems, RAG pipelines, multi-agent orchestration, and production-grade AI infrastructure on AWS and GCP.
Engineering Benchmarks
Performance metrics from production AI systems
2+Years Experience•
100K+RAG Embeddings•
30xAPI Latency Reduced•
5+LLM Providers Integrated
- API Latency Achieved
- Embeddings in Production RAG
- Latency Reduction (30s to <1s)
<1s
API Latency Achieved
100K+
Embeddings in Production RAG
30x
Latency Reduction (30s to <1s)
- Manual Work Eliminated
- CGPA in BS Software Engineering
- End-to-End AI Systems Delivered
90%+
Manual Work Eliminated
3.43
CGPA in BS Software Engineering
5+
End-to-End AI Systems Delivered
How I Build AI Systems
From prototype to production-grade infrastructure
- 1. ScopeUnderstand constraints: latency budgets, cost limits, scaling requirements, and data characteristics.
- 2. DesignChoose the right architecture: which models, embedding strategies, chunking approaches, and orchestration patterns.
- 3. BuildFast prototypes to validate before committing to production code. Iterate on retrieval quality and accuracy.
- 4. DeployShip with observability on AWS/GCP. Docker, CI/CD pipelines, and event-driven architectures.
- 5. OptimizeImprove based on production data: reduce latency, cut costs, improve retrieval accuracy and reduce hallucinations.
Featured Projects
Deep dives into products I've shipped from 0 to 1
Active
Multi-Agent AI Automation System
Orchestrated multi-agent workflows using LangGraph and LangChain for complex task execution. Integrated LLMs with external systems via APIs, messaging, and automation platforms.
Active
Document AI & OCR Pipeline
End-to-end document processing system using AWS Lambda, SQS, S3, and OCR engines. Implements pipelines for document ingestion, classification, validation, and structured data extraction.
My Journey
Building production AI systems from day one
AI Context Engineer
Xperion (Remote)
- •Designed AI-driven document processing pipelines using AWS Lambda, SQS, and event-driven architecture
- •Built workflows for document ingestion, classification, validation, and structured data extraction
- •Implemented context-aware retrieval and grounding techniques to improve output accuracy
- •Developed human-in-the-loop systems for validation, correction, and exception handling
- •Integrated AWS services including DynamoDB, S3, and OpenSearch for scalable data storage
- •Automated large-scale receipt processing workflows, significantly reducing manual effort
AI Engineer (GenAI / Backend Systems)
Texagon
- •Built a large-scale RAG system with 100K+ embeddings, reducing query latency to <30 seconds
- •Reduced API latency from ~30s to <1s through system optimization and edge deployment
- •Built and orchestrated multi-agent AI systems using LangChain and LangGraph across multiple LLM providers
- •Developed scalable backend services using FastAPI, PostgreSQL, and cloud platforms (AWS, GCP)
- •Built inference pipelines and model serving workflows using APIs and open-source LLMs (Ollama, vLLM)
- •Developed a scalable license plate recognition system supporting concurrent users with auto-scaling
Core Skills
Product leadership capabilities I bring to every team
Product Strategy
Product VisioningRoadmappingFeature PrioritizationPRD WritingMarket ResearchCompetitive AnalysisGo-to-Market PlanningBusiness Case DevelopmentOKR PlanningMVP Scoping
User & Growth
User ResearchUX StrategyA/B TestingOnboarding OptimizationRetention StrategyEngagement DesignConversion OptimizationFunnel AnalysisCohort Analysis
Stakeholder Leadership
Cross-Functional LeadershipExecutive CommunicationStakeholder ManagementTeam AlignmentSprint PlanningRoadmap ExecutionFeedback LoopsDecision Frameworks
Data & Insights
SQL AnalyticsDashboard BuildingPerformance MetricsStatistical AnalysisExperiment DesignHypothesis TestingData-Driven Decisions
AI Product Development
AI Product StrategyLLM Application DesignRAG System DesignPrompt EngineeringModel Selection StrategyAI UX PatternsResponsible AI
Tech Stack
Tools and platforms I use to ship AI products
React
Next.js
TypeScript
JavaScript
Python
Node.js
Tailwind
Vite
OpenAI
Anthropic
Gemini
Hugging Face
LangChain
Storybook
Replit
Retool
Supabase
PostgreSQL
MongoDB
Redis
Prisma
GraphQL
FastAPI
Vercel
AWS
React
Next.js
TypeScript
JavaScript
Python
Node.js
Tailwind
Vite
OpenAI
Anthropic
Gemini
Hugging Face
LangChain
Storybook
Replit
Retool
Supabase
PostgreSQL
MongoDB
Redis
Prisma
GraphQL
FastAPI
Vercel
AWS
Docker
Kubernetes
Nginx
GitHub
Git
Mixpanel
Slack
Google Analytics
Tableau
Figma
Notion
Linear
Airtable
Zapier
Stripe
Trello
Jira
Postman
TensorFlow
PyTorch
Jupyter
NumPy
Pandas
Framer
Docker
Kubernetes
Nginx
GitHub
Git
Mixpanel
Slack
Google Analytics
Tableau
Figma
Notion
Linear
Airtable
Zapier
Stripe
Trello
Jira
Postman
TensorFlow
PyTorch
Jupyter
NumPy
Pandas
Framer
AI Platforms
OpenAIAnthropicGeminiElevenLabsWhisperRunway ML
Product Analytics
MixpanelPostHogAmplitudeGoogle AnalyticsHotjarFullStory
Product Operations
NotionLinearJiraConfluenceCodaAirtableFigma
Prototyping & Testing
FigmaV0Bolt.newLovableReplitVoiceFlowUserTesting
Data & SQL
RetoolTableauRedshiftPostgreSQLSupabaseBigQuery
AI Coding Assistants
CursorClaude CodeAntigravityGitHub CopilotWindsurfV0
Let's Build Something
Looking to build production AI systems or need a GenAI engineer? Let's talk architecture.
Trusted By
Companies I've helped build and scale products
Wall of Love
Feedback from colleagues and collaborators