I build the platform under the product.
I'm Ravi Roy — an AI Platform Engineer & Infrastructure Architect. I design and ship production AI platforms, multi-tenant SaaS, agents, and the orchestration layer behind them — end to end, from Postgres to LLM routing to mobile.
- AI Platforms
- AI Infrastructure
- Workflow Automation
- Multi-Tenant SaaS
- Voice AI
- Mobile & Web
- 60+
- AI providers routed
- 10+
- Products shipped
- 7+ yrs
- Full-stack
- Multi-tenant
- By default
Not websites. Platforms, agents, and infrastructure.
Most of what I build is the hard layer underneath AI products — the orchestration, identity, and multi-tenancy that make them work at scale.
AI Platforms & Orchestration
The runtime that makes AI products possible — routing, fallback, and multi-model architecture across 60+ providers.
- AI orchestration backend (Mastra)
- Provider-agnostic model routing
- Multi-provider architecture
- Organization-aware AI memory
AI Agents & Voice AI
Agents that use tools and remember context, and voice systems that hold real conversations.
- Tool-calling agent runtimes
- Low-latency voice pipelines (STT → LLM → TTS)
- Interruption / barge-in handling
- RAG & tenant-scoped memory
Workflow Automation
Node-based pipelines and content distribution that run reliably, async, at scale.
- Node-based workflow engine
- Scheduling & async execution
- Retries & queue processing
- Multi-channel content distribution
Auth, Identity & Multi-Tenancy
The hard identity layer — solved once, reused everywhere.
- Multi-tenant orgs & projects
- RBAC, SSO, passwordless
- Audit logs & session management
- Encrypted secret vault
Mobile & Web Products
End-to-end product delivery, from data model to app store.
- React Native / Expo apps
- Next.js / React / Vue / Ionic web
- Production backends (Fastify / Express / NestJS)
- Hardware / IoT integration
Flagship work.
The platforms and products that define what I build — each a real, live system. Real screenshots, real brand, real scale.
Elevence AI
One studio for GPT-5, Claude, Gemini & 60+ models — unified behind a single interface.
VoAgents AI
AI voice agents & phone assistants that answer, schedule, and convert calls automatically.
ClipCam
Reaction clips in seconds — record over any clip, auto-stitched in the cloud.
EXL AI Playground
An interactive playground to experiment with language models, image generation, and AI tools — in real time.
Fanisin
Where every fan is in — live video, voice, and text with AI twins of creators.
HiHelloHR
A full HRMS — payroll, attendance, facial recognition, and IoT access control.
The orchestration layer behind the products.
The single most useful thing I build: the infrastructure that makes AI products possible without rebuilding the foundation every time.
- 60+
- model providers routed through one API
- 1
- provider-agnostic orchestration layer
- ∞
- fallback paths — never locked to one vendor
Model routing
Provider-agnostic router with cost-aware selection and automatic fallback across 60+ providers.
Multi-model architecture
A single request can fan out or fail over between models — no app changes required.
Organization-aware memory
Agent context isolated per tenant. Memory never leaks across organizations.
Workflow automation
An orchestration backend that chains models, tools, and steps into reliable workflows.
The platforms underneath the products.
Internal infrastructure I've built and run in production — the hard layer other products delegate to. Solved once, correctly, and reused everywhere.
AI Orchestration Platform
A provider-agnostic orchestration layer that routes every AI request across 60+ providers with fallback and cost-aware selection. Apps ask for a capability; the platform decides which model answers — so products are never locked to a single vendor.
- 60+ AI providers behind one interface
- Model routing with fallback & cost-aware selection
- Multi-provider architecture (fan-out / fail-over)
- AI workflows chaining models + tools
- Mastra backend (Bun / Hono) runtime
- Organization-aware AI memory
Workflow Orchestration Engine
A node-based workflow engine for automating multi-step pipelines — scheduled or event-driven, executed asynchronously with retries and backoff. The substrate behind automations across the platform.
- Node-based visual workflows
- Scheduling (cron + event triggers)
- Asynchronous execution
- Automatic retries with backoff
- Multi-step pipelines with branching
- Durable queue processing
Content Distribution Infrastructure
A multi-channel distribution system that schedules, repurposes, and auto-publishes content across social platforms and blogs through OAuth integrations and queue-backed processing.
- Scheduling & auto-publishing
- Content repurposing per channel
- Multi-channel distribution
- OAuth integrations
- Queue processing
Authentication Platform
The org-first identity platform every product delegates to: multi-tenant organizations and projects, RBAC, SSO, passwordless and social login, session management, audit logs, and an encrypted secret vault.
- Multi-tenant — organizations & projects
- RBAC (org / role / resource)
- SSO + passwordless + email OTP
- Social login
- Session management
- Audit logs
- Encrypted secret vault
More products.
Consumer apps, AI tools, and platforms — built on the same orchestration and identity foundation, shipped end to end.
A modern, full-stack toolkit.
From the AI layer down to the database and out to mobile — the stack I use to ship platforms end to end.
Building something that needs real infrastructure?
If you're building an AI platform, a multi-tenant SaaS, or the systems underneath one — let's talk. I take on a small number of serious projects.