Skip to content
nirav.gondaliya
OPEN TO SENIOR / FOUNDING REMOTE ROLES · AVAILABLE NOW

Nirav Gondaliya

Senior Full-Stack Engineer who builds AI infrastructure.

7+ years shipping production systems across fintech, healthtech and enterprise SaaS. I build the unglamorous infrastructure that makes AI products actually work — MCP servers, local LLM deployment, agentic pipelines, resilient data systems.

View projectsGet in touchRésumé
7+
years shipping
5M+
data points/day
8
engineers mentored
1
Lighthouse PR
ABOUT

Engineer first. Founder-shaped.

I'm a senior full-stack engineer based in India, working fully remote for the last few years. Seven-plus years of shipping production systems — most of them things real businesses pay for, monitor, and depend on every day. Core stack: TypeScript, Node.js, React, Next.js, AWS, Postgres and MongoDB.

The half of the job I love most is the AI side: agentic pipelines, multi-LLM routing, RAG over messy real-world data, on-device LLMs, MCP servers. AI-assisted development — Claude Code, Cursor, Aider — is built into the way I work, not a bolt-on. It's the reason I ship 2–3× faster than I did two years ago, and the reason I expect anyone I work with to do the same.

I've partnered directly with stakeholders in India, the US, and the UK — including enterprise teams where the C-suite was on the call. I've mentored 8 junior engineers from "help, my tests are failing" to confidently leading features. I do my best work when I own a system end-to-end, from a blank repo to production.

EXPERIENCE

Seven years in production.

Every role here shipped something users depend on — not pitch-deck prototypes.

  1. Senior Full-Stack Engineer · Ratnam Solutions

    Sep 2022 — Present · Remote
    • Built a real-time B2B analytics platform for a multinational QSR chain operating across India & Indonesia, processing 5M+ data points/day into executive dashboards (SSSG, ADS, ADT) — saved the C-suite 5–10 hrs/week.
    • Architected a Node.js + MongoDB + Redis data-warehousing pipeline; cut data-transfer latency ~40% across distributed franchise sources.
    • Primary technical liaison to the client; led requirement breakdown across a 4-engineer team.
    • Embedded AI-assisted development (Cursor, Claude Code) into the team; ~30% faster delivery on greenfield modules.
    • Owned production deploy, monitoring and incident response on AWS.

  2. Independent Consultant · Shivay Infotech

    2023 — Present (alongside full-time) · Remote
    • Built Consult Doctor — telehealth with WebRTC video consults, Razorpay payments and digital prescriptions.
    • Built the Elios GenAI Suite — image, video, TTS/STT, résumé generation and Google OAuth under one roof.
    • Mentored 8 junior engineers in full-stack development, system design and code review.

  3. Full-Stack Developer · Techniverse Software

    Jan 2022 — Aug 2022 · Remote
    • Designed serverless infrastructure (AWS Lambda, API Gateway, CloudFormation) for enterprise SaaS clients — ~35% lower infra cost vs EC2.
    • Infra-as-code cut provisioning from days to hours.

  4. Software Developer · Soft 'n' Web

    Jan 2021 — Jun 2021 · Onsite, India
    • Built conversational-AI features using Google Dialogflow and Firebase Functions — scaled to thousands of daily interactions.

  5. Software Developer · WebCodeGenie Technology

    Jan 2019 — Dec 2020 · Onsite, India
    • Promoted from intern to developer inside 6 months.
    • Shipped production CRM, ERP, Lead Management and loan systems for mid-market clients across India (Node.js, Angular, MongoDB, MySQL).
SKILLS

Things I've actually shipped with.

Grouped by where they live in my brain. Every entry is something I'd be comfortable being interviewed on or sitting down to ship in production.

Languages
  • TypeScript
  • JavaScript
  • Python
  • Kotlin
  • SQL
  • Shell
Frontend
  • React
  • Next.js (App Router)
  • Tailwind CSS
  • Framer Motion
  • Angular
Backend & APIs
  • Node.js
  • Express
  • REST
  • WebRTC
  • Webhooks
  • OAuth 2.0
  • WhatsApp Business API
AI & LLM infra
  • MCP servers
  • Multi-LLM routing
  • Agentic pipelines
  • RAG
  • On-device / local LLMs
  • Qwen 2.5
  • Cline · Aider
  • OpenAI APIs
  • Google Dialogflow
Data
  • PostgreSQL
  • MongoDB
  • Redis
  • MySQL
  • Queueing / backpressure
  • Data warehousing
Cloud & infra
  • AWS (Lambda, API Gateway, CloudFormation, Route 53, EC2)
  • Docker
  • Infra-as-code
  • Multi-tenant provisioning
  • PM2
  • UFW · SSH hardening
Payments & domain
  • PhonePe Standard Checkout v2
  • Razorpay
  • Stripe-style audit logs
  • RBAC
GPU & ML toolchain
  • NVIDIA CUDA
  • PyTorch
  • Transformers
  • Diffusers
  • Accelerate
  • FFmpeg
Practice
  • AI-assisted dev (Claude Code, Cursor, Aider)
  • Production incident response
  • Mentoring & code review
  • Cross-timezone async delivery
FEATURED WORK

Projects that paid the bills.

A sample of systems I've owned end-to-end. Real users, real money, real failure modes.

UPI MCP Server

Pairs with UPI Agent (LLM client)
The problem
UPI is India's default payment rail, but there's no way for an AI assistant to mint a valid upi:// link or a scannable QR without hitting a third-party API and leaking payee details.
What I built
An offline, zero-dependency MCP server that builds and parses UPI deep links and renders PNG QR codes locally. Three tools — create_upi_link, create_upi_qr, parse_upi_link — each with VPA validation and amount bounds. No API keys, no network calls, no NPCI affiliation: pure deterministic logic over the public UPI URI spec.
My role
Sole author. Designed the tool schema, wrote the URI builder/parser and the QR pipeline.
TypeScriptMCP SDKNode 18+QR PNGUPI URI specOffline
View repository

UPI Agent (LLM client)

Pairs with UPI MCP Server
The problem
Most agent demos hide the loop behind a framework. I wanted a minimal, readable proof that a tiny TypeScript loop + a free open-weight LLM can drive a real MCP server end-to-end.
What I built
A from-scratch agent loop that turns plain-English requests ("send ₹250 to alice@upi for lunch") into UPI links and QR codes by autonomously calling tools on the upi-mcp server. No agent framework — just history + tools → LLM → tool call → result → repeat. Runs on Groq's free Llama 3.3 70B endpoint at temperature 0 for deterministic tool dispatch.
My role
Sole author. Wrote the agent loop, the tool-dispatch layer and the MCP client wiring.
TypeScriptMCP Client SDKGroq APILlama 3.3 70BTool callingNode 18+
View repository

PhonePe MCP Server

The problem
PhonePe handles a huge slice of India's digital payments, yet there's no first-party way to plug it into Claude, agents, or any MCP-capable AI client.
What I built
An unofficial TypeScript MCP server for PhonePe Standard Checkout v2. Five tools — create_payment, get_order_status, initiate_refund, get_refund_status and a timing-safe verify_webhook_signature. OAuth tokens are fetched, cached and refreshed automatically. Sandbox and production environments. Unit-tested against a mocked fetch — no live calls.
My role
Sole author. Designed the tool schema, wrote the auth/cache layer, ship and maintain it solo.
TypeScriptMCP SDKNode 18+OAuthWebhook cryptoVitest
View repository

Multi-Store Retail ERP / CRM / PM

The problem
A multi-store retail chain was running ops across five disconnected SaaS tools, each with their own access model.
What I built
A single ERP with embedded CRM and project management. Full role-based access control down to the field level — owners, store managers, cashiers and accountants each see a different app.
My role
End-to-end: schema, RBAC engine, deploy, monitoring.
Next.jsNode.jsPostgreSQLRBACStripe-style audit log
Private · client work

Auto-Provisioning Lead Funnel

The problem
Sales team needed every qualified lead to land on a working, isolated instance of the product — not a generic demo.
What I built
A lead funnel that programmatically spins up a fresh AWS instance of the product on its own subdomain per client. New tenant, new DNS, new DB, all wired in under a minute.
My role
Architected the multi-tenant provisioning pipeline and DNS automation.
Node.jsAWS SDKRoute 53IaCPostgres
Private · client work

Resilient API → SQL Ingestion Pipeline

The problem
A fragile upstream API would buckle under volume — rate-limits, timeouts, and silent partial failures were corrupting downstream reports.
What I built
A high-volume ingestion pipeline with adaptive rate-limiting, exponential backoff, queue-based backpressure and at-least-once delivery into SQL. Upstream is now treated as the unreliable network it actually is.
My role
Designed the resilience model and rebuilt the pipeline from scratch.
Node.jsSQLQueueingBackpressureIdempotency keys
Private · client work

On-Device LLM for Android

The problem
Wanted to prove a useful LLM could run fully offline on a phone — no cloud, no API key, no telemetry.
What I built
A custom Android app that loads and runs a quantized LLM locally. Conversational UX, streaming tokens, fully offline. Inference happens on-device with model files bundled or sideloaded.
My role
Solo build — app shell, prompt loop, model integration.
AndroidKotlinLocal inferenceQuantized LLM
Private · client work

WhatsApp Appointment Booking + Payments

The problem
Clinic clients were dropping off any booking flow that required them to leave WhatsApp.
What I built
Full appointment booking inside WhatsApp via the Business API — slot picking, confirmations, reminders and an integrated payment gateway. Zero context-switch for the patient.
My role
End-to-end: conversation design, payment integration, deploy.
WhatsApp Business APINode.jsPayment gatewayWebhooks
Private · client work

Elios GenAI Suite

The problem
Small businesses wanted one place for all the GenAI tools they kept hearing about — not ten separate logins.
What I built
A unified suite: image generation, video processing, TTS, STT, AI résumé generation and Google OAuth — all under one dashboard with shared billing and history.
My role
Full-stack lead for the platform and integrations.
OpenAI APIsNext.jsNode.jsGoogle OAuthFFmpeg
Private · client work

POS Application

The problem
Retailers needed a POS that worked on cheap Android hardware, kept selling when the internet died, and synced cleanly once it came back.
What I built
Full point-of-sale system: cart, taxes, discounts, receipt printing, offline queue, and reconciliation with the central database.
My role
Built the offline-first sync layer and the reconciliation engine.
Node.jsReactOffline-first syncThermal printer
Private · client work

Consult Doctor

The problem
Independent doctors couldn't afford the per-seat pricing of enterprise telehealth platforms.
What I built
Telehealth platform with WebRTC video consultations, Razorpay payments and digital prescriptions a patient can download or forward to a pharmacy.
My role
Built the video layer and the prescription pipeline.
WebRTCNode.jsRazorpayPDF generation
Private · client work

Local Agentic Code Pipeline

The problem
Cloud-LLM dev tools are great until you're working with proprietary code you can't legally send to a third party.
What I built
A self-correcting local agent (Qwen 2.5 + Cline + Aider) that generates project structure, writes tests, runs them, reads stack traces and iterates until everything's green. Entirely local — no cloud LLM, no telemetry.
My role
Designed the self-correction loop and the prompt scaffolding.
Qwen 2.5ClineAiderLocal inferenceTest runner loop
Private · client work
MORE TOOLS

Small things I've shipped on a Sunday.

20+ tools built — these are a sample. Some scratch my own itch, some still get daily use.

Todo + scratchpad app

↗

Daily driver. Todos and free-form notes living together, keyboard first.

View repository

File diff checker

Private · client work

Drop two files, get a clean side-by-side diff in your browser.

Google Maps lead scraper

Private · client work

Pulls structured business leads from Maps for any query + radius.

Phone-number spam classifier

Private · client work

Heuristic + ML model that flags likely spam callers by pattern.

Number-based image-region highlighter

Private · client work

Annotate any image with numbered regions, exportable as overlay.

Side-by-side image review platform

Private · client work

Store uploads, admin approval queue, side-by-side comparison.

AI server provisioner (Ubuntu 24.04)

↗

One-shot Ubuntu 24.04 setup for GPU AI hosts: NVIDIA drivers + CUDA, Docker, a shared PyTorch venv (transformers, diffusers, accelerate), Node LTS + PM2, UFW. Idempotent.

View repository
OPEN SOURCE

A small dent in a very large tool.

google / lighthouse

Authored a merged PR to Google Lighthouse.

Lighthouse is the web-performance auditing tool the rest of us use to grade our sites — used by millions of developers, baked into Chrome DevTools, PageSpeed Insights, and basically every CI performance gate worth running. Contributing to it sharpened my taste for the kind of details that move real-world performance numbers (and, yes, this site is built to score against it).

Web performanceOpen sourceMerged PR
View Lighthouse repo
MENTORING & LEADERSHIP

I make the next engineer better.

8
Junior engineers mentored

One-on-one mentoring across full-stack, system design and code review. Pairing, office hours, design docs reviewed line by line. The bar I hold them to is the bar I'd want a teammate to hold me to.

4
Engineers I led as technical liaison

At Ratnam Solutions I was the bridge between the client's C-suite and a four-engineer team — breaking down vague asks into shippable scope, sequencing the work, unblocking people, owning the "is it done?" conversation.

CLIENTS & REACH

Across three continents, one keyboard.

I've delivered for clients and stakeholders across India, the United States, and the United Kingdom — from founders moving fast on a first build, to enterprise teams where the decision-maker was on a board call ten minutes after we hung up.

🇮🇳 India🇺🇸 United States🇬🇧 United KingdomEnterprise · C-suiteStartup founders
WHAT I'M LOOKING FOR

Senior or founding full-stack / AI-infrastructure roles. Remote, working with global teams. Open to full-time or contract. I do my best work owning systems end-to-end from zero.

Role
Senior · Founding · Staff
Setup
Remote, async-friendly
Shape
Full-time or contract
CONTACT

Building something from scratch?
Let's talk.

I read every cold email. The fastest way to get a real reply is to tell me what you're building, who's on the team, and what would make the next 90 days a win.

Start the conversationDownload résumé (PDF)