webZaby
Tehzim Pathan

AI-powered infrastructure for local businesses.

webZa is building an integrated platform combining specialized AI, website generation, hosting, SEO and SaaS infrastructure to help small and local businesses get online.

Researching•Training•Testing•Improving
W1
W1 MiniLMModel Card · v0.1.4
Training Active
DeveloperwebZa
StatusActive research
Parameters142M
Training tokens8.2B
Training progress48%of planned training run
Training loss2.14
GPU hours180
Latest checkpointW1-0.1.4
Training Run Progress48%

W1 is private research. Metrics shown are research checkpoints and not production guarantees. W1 is not the final model.

About

I'm Tehzim Pathan. I research, experiment and build technology for real-world businesses.

I'm currently working as a Core IT Recruiter, working closely with technology teams and professionals across the IT ecosystem. Over the years, that has given me a deep understanding of modern technology stacks — from programming languages and frameworks to databases, cloud technologies, software architecture and the different components that make real-world systems work.

I didn't start with a certificates wall — I started with products. A doctor's appointment system. A platform for housing societies. A gifting product that turns memories into songs. Websites for local food and automotive businesses. Each one began with the same question: what does this business actually need to work better?

Over time, that question became a way of working. I look at the business first — customers, leads, bookings, payments, follow-ups, operations — and then design the smallest system that genuinely solves the problem. Sometimes that's a single professional website. Sometimes it's a full multi-tenant SaaS platform. The scope follows the problem, not the other way around.

My IT background helps me look beyond the interface. I understand how the pieces connect — frontend, backend, databases, APIs, infrastructure, hosting, authentication, automation and AI — which allows me to think about products as complete systems rather than just websites.

I work across the full journey: understanding, planning, design, build and launch. When a product goes live, it goes live ready for real users — hosting configured, domain connected, workflows tested. That end-to-end involvement is the point: you're not piecing together freelancers, you're working directly with the person behind Webza's research and products.

Building toward a bigger idea

My ambition with Webza goes beyond building individual websites.

I'm working toward an AI-powered Website-as-a-Service (WaaS) platform designed specifically for local businesses — making it easier and more affordable for businesses to establish, manage and grow their online presence.

As part of that journey, I'm researching and training specialized AI models designed to understand real businesses, their services, industries, content and website requirements.

W1 MiniLM is Webza's current private AI research initiative — exploring how smaller, specialized models can support website generation, modification, SEO, business content and Webza's internal workflows.

The research is ongoing. I'm continuously experimenting with models, training approaches, website structures, business terminology and real-world website requirements to understand how specialized AI can become more useful for businesses.

What drives me is simple: I enjoy turning business ideas into things that exist — products people use, systems businesses rely on, and technology that makes it easier for businesses to move forward.

And Webza is where I'm bringing that ambition together.

See my work

One builder, the full range.

Website development

Business websites engineered around leads and customer trust.

SaaS products

Multi-tenant platforms with dashboards, portals and subscriptions.

AI solutions

AI applied to real workflows — generation, support, automation.

Automation

Repetitive business work handed to reliable software.

Business systems

Management systems for the daily operations of a business.

Digital products

Ideas productized into things people actually pay for.

W1 MiniLM – Training Focus

What W1 is being trained to do.

Twelve focused capabilities W1 is being researched and trained on — each one targeted at a specific business website workflow.

01

HTML

Generating semantic, accessible and search-friendly HTML structures for business websites.

Foundation markup generation
02

CSS

Producing responsive, mobile-ready visual styling that matches the business brand.

Visual styling & layout
03

JavaScript

Adding lightweight interactivity, validation and behaviour where business pages need it.

Client-side interactivity
04

React

Composing component-based website structures that map to real business sections.

Component composition
05

Tailwind

Generating utility-first styling for fast, consistent and maintainable websites.

Utility-first styling
06

SEO

Writing search-friendly page titles, meta descriptions, headings and content structures.

On-page SEO generation
07

Schema.org

Generating structured-data markup so search engines understand local businesses.

Structured data generation
08

Landing Pages

Designing focused, conversion-friendly landing pages for services and campaigns.

Conversion-focused layouts
09

Business Website Content

Writing hero, services, gallery, about and contact copy that fits Indian SMBs.

Business copywriting
10

Website Modification

Editing existing pages — updating services, gallery, hours and content over time.

Iterative editing
11

Indian SMB Terminology

Understanding how Indian small businesses describe themselves, their services and customers.

Domain vocabulary
12

webZa Internal AI Workflows

Optimizing W1 to operate inside webZa's internal research and generation pipeline.

Internal workflow fit
W1 MiniLM – Scope

W1 research focus.

What W1 is — and isn't — being researched to do.

Research Focus

HTMLCSSJavaScriptSEOSchemaBusiness CopyIndustry ContextWebsite Modification

Technology Focus

HTMLCSSJavaScriptReactTailwindSEOSchema.org

Domain Knowledge

RestaurantsSalonsClinicsGymsReal EstateAutomotiveLocal Services
Also exploring:
CafésBakeriesPhotographersInterior designersTravel businessesProfessional services

W1 is intentionally narrow in scope. We are not researching it for general-purpose text generation, image generation or code unrelated to business websites.

The Approach

Why are we researching smaller specialized models?

General-purpose AI models are powerful, but they try to do everything for everyone. For business website generation, that broadness comes with cost, latency and a higher chance of generic, off-brand output.

webZa is taking a different research direction: smaller, focused models trained specifically on the workflows of building, modifying and improving Indian small business websites — HTML, CSS, JavaScript, SEO, schema and business copy.

The trade-off is real. Specialized models cover less surface area — but on the surface they cover, they can be sharper, cheaper to run and easier to improve.

We are not trying to build another general-purpose AI model.W1 is being researched for one focused job: business website generation for Indian small businesses.

General-purpose AI

  • Trained on broad, mixed data
  • Tries to handle many tasks
  • Larger, costlier to run
  • Generic, off-brand output risk

webZa's Approach — Specialized Models

  • Trained on business-website workflows
  • Focused on one job, done well
  • Smaller, cheaper to run
  • Sharper, on-brand website output
Active Development

W1 is still being trained.

W1 is not the final model. webZa continuously experiments with training data, website structures, business terminology, generation strategies and evaluation methods. Each improvement cycle feeds the next.

The W1 Improvement Loop

01Research
02Train
03Generate
04Evaluate
05Test on real websites
06Refine
07Retrain
08Improve

Loops back · Step 08 → Step 01

Under Research

Evaluation-driven improvement.

webZa is researching reinforcement and preference-based learning techniques for improving specialized website-generation models. The idea is to use evaluation signals — human and automated — to steer W1 toward better website output over time.

This is a research direction. W1 does not currently use reinforcement learning. We are exploring how evaluation signals could improve specialized models in the future.

* Conceptual workflow. Not currently implemented in W1.

Preference Learning FlowResearch
BusinessL0
W1L1
Solution A · B · CL2
Human / automated evaluationL3
Preferred outputL4
Training signalL5
Model improvementL6
—— Exploring the Solution Space

One business. Multiple solutions.

The same business can have multiple valid website strategies depending on its customers, services, positioning and goals.

Business BriefPremium unisex salon targeting young professionals.
Solution A

Premium

Brand-led layout with elegant typography, large imagery and a refined tone.

HeroStoryServicesGalleryContact
Solution B

Local Conversion

Conversion-focused layout with clear CTAs, services, reviews and booking.

Hero + CTAServicesReviewsBookingLocation
Solution C

Social / Visual

Visual-first layout with photo grids, Instagram embeds and social proof.

GalleryFeedStoriesAboutFollow

Conceptual example of the W1 research workflow. This illustrates the kind of multi-solution exploration W1 is being researched for — not currently produced solutions.

From Model to Website

From model research to live websites.

The end-to-end pipeline W1 research feeds into — from business information all the way to a deployed website.

01Business information
02Industry context
03W1 processing
04Website structure
05HTML generation
06CSS + Tailwind
07SEO + Schema.org
08Business copy
09Deployment
10Live website
Live domain examplesgatexa.inwarismeats.com
Research in Progress

AI-Powered SEO.

W1 is being researched to generate search-friendly SEO output across every layer of a business website — from keywords to structured data to internal linking.

Keywords
Page titles
Meta descriptions
Headings
Schema.org
Local-business content
Service pages
Landing pages
Internal linking
Search-friendly content structure

* Research in progress — no rankings or traffic are guaranteed.

Evaluation Environment

Real websites. Real testing.

W1 research is tested against real, live business websites — not just synthetic examples. Real businesses surface real edge cases.

RestaurantsSalonsClinicsGymsReal EstateAutomotiveProfessional ServicesLocal Businesses
01
Live business website

Tested against W1 research

02
Live business website

Tested against W1 research

03
Live business website

Tested against W1 research

04
Live business website

Tested against W1 research

05
Live business website

Tested against W1 research

06
Live business website

Tested against W1 research

* Real-world testing is part of the W1 research loop. Examples shown are illustrative — actual live websites are listed on the home page.

Range of Work

Built around real business problems.

Real-world systems and websites webZa has worked on — the same kind of business context W1 research draws from.

01
apointo.in

APOINTO

Healthcare & Clinics

02
gatexa.in

GATEXA

Society Management

03
feelsay.in

FEELSAY

Ecommerce & Gifting

04
warismeats.com

WARIS MEATS

Local Food Business

05
QADRI CAR SCRAP

QADRI CAR SCRAP

Automotive & Scrap

06
TAPNEST

TAPNEST

White-label SaaS

07
SEATO

SEATO

Booking & Mobility

08
AI WEBSITE SERVICE

AI WEBSITE SERVICE

AI-powered Systems

09
US RETAIL POS

US RETAIL POS

Retail Operations

10
APOINTO · GATEXA

APOINTO · GATEXA

Business Management

Research in Progress

W1 is research in progress.

W1 is not the final model. webZa continues to research, train, evaluate and improve specialized models for business website generation.

W1Made with Webza W1