Vedansh Vijayvargia

Full Stack AI Developer

Web apps, AI agents and business software — built end to end.

Start a project

Hiring? Resume (PDF)

Scroll

// skills

The stack I ship with.

AI & LLMs

Voice agents, chatbots and LLM features running in production.

OpenAI API
Anthropic Claude
Retell AI voice agents
LangChain
RAG
Hugging Face
Fine-tuning (LoRA)
Ollama

Frontend

Dashboards, admin panels and fast public websites.

Next.js
React
TypeScript
JavaScript (ES6+)
HTML & CSS
Bootstrap
AJAX

Backend

APIs, business logic and background jobs that hold up under real users.

Node.js
PHP (OOP)
REST APIs
Express
Python
Auth & roles
Redis & queues

Data & cloud

Databases, deployment and keeping things fast once they’re live.

MySQL
PostgreSQL
AWS
MongoDB
Docker
Linux
cPanel / WHM
Git

// selected work · 2023–2026

Things I’ve built that people use every day.

Live products and client sites, each built end to end. Open any of them right here.

01 · Product

SmartChase

An AI reminder system that does the chasing so accountants don’t have to.

Node.js
TypeScript
Retell AI
LLM prompts

Case study

02 · Freelance · Australia

bim365.in

A full BIM services platform for architecture, engineering and construction teams in Australia.

PHP
MySQL
Blog / CMS
Local SEO
Lead forms

Case study

03 · Freelance · BIM

bimcrafter.com

A second BIM services website, built on what worked for bim365.

PHP
Responsive
SEO
Enquiry forms

Case study

04 · Freelance · Jaipur

thewellnesscode.in

The website for Dr. Shivangi Jain’s doctor-led nutrition and wellness practice.

PHP
HTML / CSS
Blog & videos
WhatsApp booking

Case study

05 · Freelance · Hong Kong

sourcyn.com

A B2B platform, built and deployed end to end.

PHP
JavaScript
MySQL
SEO

Case study

06 · Freelance · E-commerce

shskates.com

An online store for a skate shop that runs without babysitting.

PHP
MySQL
Cart & checkout
Admin panel

Case study

07 · Freelance · Gujarat

greenvalleysolar.in

A solar company site with an admin panel the team runs themselves.

PHP
MySQL
cPanel
SSL & backups

Case study

smartchase · internal 01 / 07

Email reminderday 0

SMS follow-upday 3

AI voice call · Retellday 6

Hand-off to a humanonly if needed

Each step stops the moment the client responds.

More work

8 case studies · problem, build, result View more projects

// from clients

In their words.

Four of the people I’ve built for, on what they actually got.

bim365.in01 / 04

Vedansh built the whole bim365.in presence — service pages, city SEO, blog, testimonials and consultation forms. It’s a proper multi-page product, not a template swap. Mobile speed mattered because our clients open it from site offices.

Naman VijayFounder, BIM365 · Jaipur

sourcyn.com02 / 04

About a month from brief to live — product pages, enquiry form, mobile layout. SEO picked up slowly, as it does, but inbound leads are clearly up compared to our old site.

Bobby BarotFounder, Sourcyn · Hong Kong

shskates.com03 / 04

The skate shop site handles catalogue, cart and checkout without us babysitting it. Mobile performance was the main ask — that part delivered.

Anshul VijayOwner, SH Skates · Bhilwara

mantriitsolutions.com04 / 04

mantriitsolutions.com explains our services without clutter. Enquiries come through one form instead of scattered DMs.

Akshay MantriFounder, Mantri IT Solutions · Kota

// free audit

Get a free website audit.

Fill in the form and I’ll send you the three fixes that would bring you more enquiries, within one business day. Free, no obligation.

Speed on mobile data
Mobile layout
Google basics
How easy it is to contact you

// how a build actually goes

Most of my work starts with someone doing the same boring thing every day.

01

Find the step that eats the week

Chasing unpaid invoices. Re-typing orders into a spreadsheet. Setting up the same voice agent by hand for every client. I start by watching how it's done today.
02

Write it as plain, readable code

Small functions, obvious names, no clever tricks. If the next developer can't follow it in five minutes, it isn't finished.
03

Ship it, then watch it run

Tested in staging, deployed properly, and checked against real usage — not just the happy path from the demo.

The code on the right is the core idea behind SmartChase, a reminder system I built: email, then SMS, then an AI voice call, and only then a human.

escalate.ts channels.ts smartchase / src

// escalate a reminder until someone responds
const ladder = ['email', 'sms', 'voice'] as const;
 
export async function chase(invoice: Invoice) {
  for (const channel of ladder) {
    await send(channel, invoice.client);
    const reply = await waitForReply(invoice, days(3));
    if (reply) return close(invoice, reply);
  }
  // nobody answered: hand it to a person
  return assignToHuman(invoice);
}

$ npm test

✓ escalates email → sms → voice

✓ stops as soon as the client replies

✓ hands off to a human at the end

$ git push origin main

→ deployed · running

TypeScriptLn 1, Col 1● build

EmailReminder sent · no reply after 3 days

SMSFollow-up sent · still quiet

AI callClient picked up · payment promised

ClosedNo one on the team had to chase it

// about

The developer who turns the demo into the real thing.

I’m Vedansh, a full-stack developer in Vadodara. For three years I’ve built the parts people actually depend on — ERP screens thousands of employees log into, client sites that load on a site-office phone, and AI agents that make the follow-up call so nobody on the team has to.

Right now I’m at Technosmart AI, where I built the dashboard our team uses to set up and manage AI voice agents for every client. Before that I spent almost two years at PayGURU Technosoft on an ERP platform used by companies like Gujarat Fluorochemicals and Lanxess — which is where I learned that a slow query is a real person waiting.

Outside the day job I take on a few freelance projects at a time: scoped in writing, built in the open, and handed over so you’re never stuck with me.

Start a conversation

~/vedanshonline

$ whoami --json
{
  "name": "Vedansh Vijayvargia",
  "role": "Full Stack · AI/ML",
  "now": "Technosmart AI",
  "based": "Vadodara, IN",
  "years_shipping": 3,
  "stack": ["Node", "Next", "PHP", "Python"],
  "bugs_fixed": "lost count",
  "open_to": "freelance"
}
$

Node.jsNext.jsPHPPythonRetell AIMySQL

Based in: Vadodara, Gujarat · remote with clients
Currently: Full Stack Developer, Technosmart AI
Education: B.Tech CSE (AI), Parul University, 2024
Certification: Full Stack Web Dev Bootcamp, Coding Ninjas
Languages: English, Hindi

// in production

What the work did once real people used it.

Figures from the PayGURU ERP platform, where I owned modules from schema to screen.

9,000+

people log into the ERP platform I engineered

Across 5+ client companies, including Gujarat Fluorochemicals and Lanxess.

~40%

less query time

Rewriting and indexing the SQL behind the slowest screens.

~80%

faster page loads

Same modules, same users — just less waiting.

10+

ERP modules delivered

HRMS, finance, inventory and the pieces that connect them.

30+

JasperReports built

Report prep time cut by about half; no more Friday exports to Excel.

// experience

The journey so far.

Scroll and the pen writes it out — from the AI platform I work on today back to where it started.

Nov 2025 — now

Full Stack Developer Technosmart AI · Vadodara, onsite

Built an internal dashboard for configuring and managing Retell AI voice agents across clients, replacing manual per-agent setup.
Integrated LLM-backed voice APIs to run automated inbound support agents with per-client prompt configuration.
Backend in Node.js and TypeScript, front-end in Next.js — non-technical staff now create agents and manage client setups themselves.

Jan 2024 — Nov 2025

Full Stack Developer PayGURU Technosoft · Vadodara, onsite

Engineered an ERP platform in OOP PHP and JavaScript serving 9,000+ users across 5+ client companies, including Gujarat Fluorochemicals and Lanxess.
Delivered 10+ modules (HRMS, finance, inventory) and optimised SQL, cutting query time by ~40% and page load by ~80%.
Built real-time AJAX features and 30+ JasperReports, lifting engagement ~25% and cutting report-prep time ~50%.

Feb 2023 — Dec 2023

Web Developer Blockverse Infotech · Jaipur, remote

Built responsive PHP/MySQL websites with interactive JavaScript components, delivered with full client sign-off.
Form validation and dynamic content loading; managed cPanel hosting, FTP, databases and on-page SEO.

2023 — now

Freelance Coders Universe · alongside full-time work

bim365.in, bimcrafter.com, thewellnesscode.in, sourcyn.com, shskates.com, greenvalleysolar.in and mantriitsolutions.com — each scoped, built, deployed and handed over by me.

2024

B.Tech, Computer Science (AI) Parul University · CGPA 7.48

// ai lab

Below the API call.

At Technosmart AI I ship voice agents to production. Alongside that, I’ve been fine-tuning small open-weight models — Gemma 2B and 7B, Mistral 7B — to understand what actually happens when a model learns something new. This is the loop I run.

Voice agents

Inbound support lines on Retell AI with per-client prompts, set up from a dashboard instead of by hand. Live at Technosmart AI.

production

Workflow agents

Follow-ups that escalate on their own, like SmartChase. A person steps in only when the automation runs out of road.

production

Control dashboards

Next.js panels where non-technical staff edit prompts, switch flows on and off and manage clients without a deploy.

production

The fine-tuning loop

Seven steps, from raw conversations to a model answering inside an app. Scroll to run it.

01
Collect and clean the data
Real conversations are messy. Dedupe, fix encodings, throw out the junk. Most of the quality is decided here.
hands-on
02
Anonymise it
Names, numbers and anything personal get stripped or replaced before a single training step runs.
hands-on
03
JSONL, then tokenise
One example per line in the chat format the model expects — then check how the tokenizer splits it, because tokens drive cost, context and memory.
hands-on
04
Configure LoRA / QLoRA
Learning rate, batch size, gradient accumulation and epochs, chosen around what the GPU can hold. A 4-bit base is what makes a 7B model fit at all.
experimenting
05
Watch the loss, then evaluate
Training and validation loss side by side, then the tuned model against the base on held-out prompts. Lower loss alone doesn’t mean better answers.
experimenting
06
Save, quantise, serve
Keep or merge the adapter, reload it to be sure it works, quantise for inference and run it locally through Ollama.
experimenting
07
Wire it into an app
The part I do every day: model inference behind an API a product can depend on, with timeouts, fallbacks and logs.
production

Want to see the production side? The assistant on this site is an LLM wired into a small PHP proxy. Ask it anything about my work.

Talk about an AI project

// writing

Notes from real projects.

Longer write-ups on things I’ve had to figure out on the job — ERP query tuning, Retell voice agents, RAG versus fine-tuning, PHP deploys.

All articles →

AI / Machine LearningMay 2026

RAG vs Fine-tuning: How I Actually Decide for Production AI Apps

A practical decision framework from a working AI developer — when retrieval beats fine-tuning, when it does not, and the cost numbers nobody puts in their slides.

Read article

AI / Machine LearningMay 2026

How I Cut a Client's OpenAI Bill by 62% Without Hurting Quality

Eight techniques that reduced a production AI app's monthly inference bill from $1,470 to $558 — model routing, prompt diet, caching, and the boring infra wins nobody talks about.

Read article

AI / Full StackApril 2026

Building a Production AI Dashboard with Next.js, Node, and Streaming LLMs

A field guide from the dashboard I ship at Technosmart — App Router, server actions, SSE streaming, Retell webhooks, and the small architecture choices that decide whether your AI product feels real.

Read article

AI / Machine LearningApril 2026

Claude vs GPT-4 in Production: A Working Developer's Take

I ship features on both. Here is how I actually choose between Claude and GPT-4 for real products — broken down by task, latency, tool use, and the small behaviours that decide it.

Read article

AI IntegrationMarch 2026

Integrating Retell AI for Voice Ops: A Production Walkthrough

Building voice agents that actually answer real customer calls — webhook handling, prompt design, escalation logic, and the latency tricks that keep conversations feeling human.

Read article

// contact

Got a workflow that’s eating your week?

Tell me what you’re building and roughly what you want to spend. I’ll reply within one business day with what’s realistic — and if I’m not the right fit, I’ll say so.

Taking on a small number of freelance projects alongside my role at Technosmart AI.

What do you need? Website Web app or dashboard AI agent or automation ERP or internal tool Not sure yet

Your name

Rough budget (optional)

When do you need it? (optional)

Current website (optional)

What are you trying to get done?

By sending this you agree to the privacy policy and terms. Prefer chat? Message me on WhatsApp.

Full Stack AI Developer

AI & LLMs

Frontend

Backend

Data & cloud

SmartChase

bim365.in

bimcrafter.com

thewellnesscode.in

sourcyn.com

shskates.com

greenvalleysolar.in

More work

Get a free website audit.

Most of my work starts with someone doing the same boring thing every day.

Find the step that eats the week

Write it as plain, readable code

Ship it, then watch it run

The developer who turns the demo into the real thing.

people log into the ERP platform I engineered

less query time

faster page loads

ERP modules delivered

JasperReports built

Full Stack Developer Technosmart AI · Vadodara, onsite

Full Stack Developer PayGURU Technosoft · Vadodara, onsite

Web Developer Blockverse Infotech · Jaipur, remote

Freelance Coders Universe · alongside full-time work

B.Tech, Computer Science (AI) Parul University · CGPA 7.48

Voice agents

Workflow agents

Control dashboards

The fine-tuning loop

Collect and clean the data

Anonymise it

JSONL, then tokenise

Configure LoRA / QLoRA

Watch the loss, then evaluate

Save, quantise, serve

Wire it into an app

RAG vs Fine-tuning: How I Actually Decide for Production AI Apps

How I Cut a Client's OpenAI Bill by 62% Without Hurting Quality

Building a Production AI Dashboard with Next.js, Node, and Streaming LLMs

Claude vs GPT-4 in Production: A Working Developer's Take

Integrating Retell AI for Voice Ops: A Production Walkthrough

Got a workflow that’s eating your week?