Currently: building product features at Userflow

Full Stack
Developer

SDE 2 at Userflow, where I work across the stack: Elixir/Phoenix and GraphQL services, React and TypeScript on the frontend, and LLM features like SmartFlow. EEE → SWE by sheer stubbornness.

React TypeScript Node.js Elixir/Phoenix PostgreSQL
View Experience Get in touch
developer.ts 
// ~ SDE 2
const developer = {
  name:     "Vijayaraagavan S",
  alias:    "Vijay",
  role:     "SDE 2",
  location: "Salem, India",

  currently: {
    company: "Userflow",
    focus:   ["full stack features", "LLM features"],
  },

  stack: {
    frontend: ["React", "TypeScript", "Vue.js"],
    backend:  ["Elixir", "Phoenix", "Node.js", "GraphQL"],
    data:     ["PostgreSQL", "ClickHouse"],
    devops:   ["Docker", "K8s", "CircleCI"],
  },

  contact: () => "svijayaraagavan@gmail.com",
};

// about

I'm a full stack developer. At Userflow I work across the Elixir/Phoenix and GraphQL services and the React app and embeddable SDK, where I built features like confetti, reactions and comments. Before that, at Ardhika, I built an offline-first restaurant POS in Elixir/Phoenix and Vue.js serving 1000+ concurrent users.

I also bring LLM features into the product. I shipped SmartFlow end to end: it turns a recorded browser session into a publishable product tour, and every generated flow is validated through the production changeset, so the model cannot emit something broken. I also own the OAuth and RBAC layer of Userflow's MCP server.

Before all this: a degree in Electrical and Electronics Engineering from the College of Engineering Guindy, Anna University, then teaching myself software engineering through personal projects and relentless curiosity.

// experience

Userflow

current

SDE 2

Jun 2024 – Present

Coimbatore, India

  • ▸ Build features across the stack, from Elixir/Phoenix and GraphQL services to the React frontends of the Userflow app and its embeddable end-user SDK
  • ▸ Built end-user features in the embeddable React SDK that runs inside customers' web apps, including confetti, reactions and comments
  • ▸ Built SmartFlow end to end, an LLM feature that turns a recorded browser session into a publishable product tour, with every generated flow validated through the production changeset so an invalid tour cannot ship
  • ▸ Created 18 of the 43 tools in Userflow's MCP server and own its OAuth 2.1 authorization layer: dynamic client registration, RFC 9728 metadata and per-scope RBAC
  • ▸ Built LLM cost and abuse controls: company-level spend caps, per-feature hourly quotas, timeout cancellation, and token accounting that attributes reasoning and cached-token spend
  • ▸ Built the evaluation programme for an LLM onboarding-planning pipeline: golden set, budget-capped LLM-as-judge, weighted scoring rubric, and a regression suite for ~20 provider failure modes
React TypeScript Elixir Phoenix GraphQL PostgreSQL ClickHouse Oban OpenAI API Gemini MCP Python

Ardhika Software Technologies

Software Engineer

May 2022 – Jun 2024

Chennai, India

  • ▸ Built an offline-first restaurant POS end to end (Elixir/Phoenix backend, Vue.js frontend) serving 1000+ concurrent users
  • ▸ Implemented real-time sync via WebSockets handling 1000+ transactions/hour at 99.9% reliability
  • ▸ Designed microservice architecture and optimized PostgreSQL schemas, cutting query response time by 20%
  • ▸ Developed custom Android plugins (Java) bridging the web app with native hardware (printers, scanners)
React Vue.js Node.js Elixir Phoenix PostgreSQL WebSockets Golang Java Docker Jenkins

// skills

Frontend

React TypeScript JavaScript Vue.js

Backend

Elixir Phoenix Node.js Fastify GraphQL Ecto ORM Oban Golang

Databases

PostgreSQL ClickHouse MongoDB Firebase

AI / LLM

LLM pipelines Structured outputs Tool calling / MCP Evals & LLM-as-judge RAG

DevOps & Cloud

Docker Kubernetes Terraform CircleCI Jenkins Cloudflare Linux

Other languages

Python Java Dart

// projects

Multi-Tenant Workspace Apps

personal

Private multi-tenant platform where organizations build and run apps in workspaces, with an AI agent that builds apps in place. One of its core engineers, on the tenant data, orchestration and infrastructure layers.

  • ▸ Built unidb, a Go service designed to serve thousands of tenant SQLite databases per node over the Hrana (libSQL) protocol, with WAL shipping to object storage, sync/async replication and per-database CPU, size and rate quotas
  • ▸ Built Setpoint, a Go reconciliation engine that continuously converges per-tenant resources (databases, migrations, sandbox templates) to their declared state
  • ▸ Built the infrastructure as code: Nomad, Consul and Traefik on Azure with OpenTofu, content-addressed images and Woodpecker CI, plus microVM sandboxes for agent builds
Go TypeScript Node.js Fastify React SQLite PostgreSQL Nomad Consul OpenTofu Azure

Quic Switch (open source soon)

personal

Local-first macOS desktop app that keeps project context for developers and AI coding agents, so switching projects loses nothing. Rust core with a CLI and an MCP server, a Tauri desktop app in TypeScript, and agent-defined Applets whose code runs sandboxed (QuickJS on wasmtime) behind per-action grants.

Rust Tauri TypeScript Preact MCP WebAssembly

Computer Vision Pipeline

personal

Automated image segmentation with SAM2 and YOLO to produce ML training data. Processed 10k+ images at 85% detection accuracy, cutting manual labeling by 80%.

Python SAM2 YOLO TensorFlow

// contact

Let's work together

Open to full stack developer roles. Drop me a line and let's chat.

LinkedIn GitHub