Back to Projects

Zaytri — AI-Powered Social Media Automation Platform

Founder, Product Architect & Full-Stack AI Engineer · 2026 – Present

Zaytri is a fully autonomous AI-powered social media automation platform for businesses, creators, and agencies. Users interact through natural language (text, voice, image) while AI agents handle content generation, review, scheduling, optimization, and analytics. Built solo end-to-end: product strategy, frontend, backend, AI orchestration, auth, database design, infra, and CLI tooling.

Key Metrics & Outcomes

  • Master Agent with 18 business intents + conversation memory
  • 5 specialized agents: Content, Hashtag, Review, Engagement, Analytics
  • 5 LLM providers with dynamic routing & fallback chains
  • ~40% AI cost reduction via Ollama local inference
  • Sub-2s semantic retrieval with pgvector RAG
  • 20+ auth endpoints: OTP, TOTP 2FA, OAuth ×4, rate limiting
  • 7-service production stack via Docker Compose + custom CLI

Project Overview

Situation

Social media teams juggle disconnected tools for content creation, scheduling, engagement, and analytics. Existing automation is rule-based — no contextual intelligence, no brand awareness, and single-LLM dependency drives up cost and fragility.

Task

Design and build a solo, production-grade multi-agent platform: natural-language control plane, specialized AI agents, brand-aware retrieval, multi-provider LLM routing, enterprise auth, and async job processing — all deployable as one stack.

Action

Architected a Master Agent orchestrator (18 intents, memory, routing to specialized agents); built a provider-agnostic LLM brain with cost/speed-based routing and failover across OpenAI, Claude, Gemini, Ollama, Groq; implemented RAG over PostgreSQL + pgvector for brand-aware generation; built Celery + Redis async pipelines with Celery Beat scheduling; shipped a Next.js 16 dashboard with server components; implemented 20+ auth endpoints (OTP, TOTP 2FA, OAuth, bcrypt, rate limiting); containerized all 7 services behind Nginx with a custom zaytri CLI.

Result

Complete AI SaaS built solo: ~40% inference cost reduction via local-model routing, sub-2s RAG retrieval, dashboard load <500ms, p95 LLM latency <3s, zero provider lock-in, and one-command platform management.

Implementation 2

Situation

Coordinating multiple LLM providers, keeping generated content brand-consistent, and keeping a real-time observability dashboard performant under async load required careful systems design.

Task

Ensure consistent routing logic across providers, maintain retrieval quality, and scale the dashboard without impacting backend throughput.

Action

Designed an abstraction layer for multi-LLM routing with fallbacks and cost-based selection; tuned pgvector indexes and RAG prompts for brand consistency; used Celery task chains and Redis for observability metadata; built the dashboard with server components and efficient polling for live metrics.

Result

Multi-LLM routing at p95 <3s with zero provider lock-in; measurable RAG relevance improvement; dashboard TTI <1s with no increase in backend error rate.

Technology Stack

  • Next.js 16
  • TypeScript
  • Tailwind CSS
  • FastAPI
  • Python 3.11
  • LangChain
  • Celery
  • Redis
  • PostgreSQL
  • pgvector
  • OpenAI
  • Claude
  • Gemini
  • Ollama
  • Groq
  • Docker Compose
  • Nginx
View Code