Bangladesh has over 650,000 active freelancers on global platforms — the second-largest freelancer workforce in the world. A significant portion work as graphic designers, video editors, and creative professionals. Yet every major AI creative tool was priced in USD, built without Bangla support, and completely disconnected from the Bangladeshi market.

When we started building Bottola AI, we asked a simple question: why does a designer in Dhaka have to pay $20/month in foreign currency to access the same tools a designer in New York uses — when the average monthly digital subscription budget in Bangladesh is a fraction of that? The answer wasn't technical complexity. It was that nobody had tried to build it here.

What Bottola AI Is

Bottola AI is Bangladesh's first home-grown AI content creation platform — not a localized wrapper around foreign APIs, but a platform with in-house AI models trained and fine-tuned by the Everest IT engineering team in Dhaka. The platform gives creators three core capabilities:

• AI Image Generation — Multiple in-house diffusion models supporting photorealistic images, digital art, product photography, and concept art. Prompts work in both Bangla and English.
• AI Video Generation — Text-to-video and image-to-video generation, enabling creators to produce cinematic short clips, product demos, and social content without a production budget.
• AI Music Composition — AI-generated royalty-free music tracks for videos and creative projects, commercially licensed on all paid plans.

Priced for Bangladesh

Every plan is denominated in BDT. The Go, Plus, and Pro tiers operate on a credit-based model — pay for what you create. Payment works natively through bKash, Nagad, Rocket, and standard cards. We didn't just localize pricing — we built the full payment infrastructure so any Bangladeshi with a smartphone can access tools previously reserved for people with international payment cards.

Bangla Prompt Support

Our in-house Bangla NLP engine means creators can describe their vision in their native language. The system understands Bangladeshi cultural references, traditional aesthetics, and local context that global models miss entirely. Write your prompt in Bangla — and Bottola creates exactly that.

Why This Matters

Bangladesh's creative economy is growing rapidly — social media marketing, digital advertising, e-commerce product photography, content creation. All of these require professional-grade visuals that small businesses and individual creators couldn't previously afford to produce. Bottola AI changes that math. Visit bottola.ai to start creating.

Bengali is spoken by more than 300 million people worldwide — making it the 7th most spoken language on Earth. Yet in the world of AI, it remains one of the most underserved languages. Most large language models were trained on datasets that are 90%+ English, with Bengali receiving less than 0.1% of representation.

For us at Everest IT, this wasn't an abstract research problem. It was a business-critical gap we had to solve in-house, because you simply cannot build useful AI products for Bangladesh if the underlying models don't understand Bangladeshi language, culture, or context.

The Challenge of Bangla NLP

Building good NLP for Bengali is harder than most people expect. The challenges are layered:

• Script complexity — Bangla script has 50 base characters, complex conjuncts, and vowel diacritics that significantly increase vocabulary space compared to Latin-script languages.
• Dialect diversity — Standard Bangla (Shuddho Bangla) differs substantially from colloquial Dhaka dialect, Sylheti, Chittagonian, and dozens of regional varieties. A model trained only on formal news text fails badly on everyday conversation.
• Banglish — Bangladeshis frequently write in Roman-script Bengali, mixing Bangla and English in a single sentence. This is ubiquitous on WhatsApp, Messenger, and social media. Most models have no framework for handling it.
• Low-resource data — High-quality annotated Bengali datasets are scarce compared to English, making it harder to train and evaluate models with accuracy.

How We Approach It

Everest IT's Bangla NLP is built in-house and designed specifically for the conversational, informal Bangla that actual Bangladeshi users write every day — not the formal prose found in news archives or academic papers. Our second-generation model (deployed in Q4 2025) significantly improved performance on dialect recognition, Banglish handling, and intent classification for customer service scenarios.

Both Bottola AI's prompt understanding engine and DeshiBot's conversation system run on this foundation.

Why Owning It Matters

Many companies would simply prepend "translate this to Bangla" to an API call. We deliberately chose not to. Owning the NLP stack means we control quality, latency, cost, and capability roadmap. Our products keep improving as our models improve. And Bangladeshi AI doesn't depend on a foreign company's pricing decisions or API availability.

A mid-size e-commerce brand in Dhaka receives 400 customer messages per day across WhatsApp and Facebook Messenger. Sixty percent are "Where is my order?" Each reply takes 3 to 5 minutes. That's up to 20 hours of customer service time daily — at a cost that doesn't scale with the business.

This is the core problem DeshiBot was built to solve.

The State of Customer Service Automation in Bangladesh

Most off-the-shelf chatbot platforms are built for English-speaking markets. Deploy them in Bangladesh and you immediately hit three problems:

• They don't understand Bangla — a customer typing in Bengali gets either no response or a generic error.
• They have no local context — no awareness of bKash, no knowledge of local delivery partner timelines, no understanding of how Bangladeshis actually write.
• They're priced in USD — making them inaccessible to the SMEs that need automation most.

What DeshiBot Does Differently

DeshiBot is built on Everest IT's in-house Bangla NLP engine. It understands standard Bangla, colloquial Dhaka dialect, Banglish, and informal text. When a customer sends a message, DeshiBot correctly interprets intent — order tracking, product inquiry, complaint, appointment booking — regardless of how it's phrased.

The platform integrates natively with WhatsApp Business API, Facebook Messenger, and web chat widgets. For most businesses, all three channels can be live within days of onboarding.

Where It's Working

• Retail & e-commerce — Order status, return processing, product availability, delivery tracking — automated end-to-end, with zero human involvement for the majority of cases.
• Healthcare — Appointment booking, doctor availability, and follow-up reminders via WhatsApp. Clinics report significantly reduced front-desk call volume.
• Finance — Account balance queries, transaction history, loan eligibility screening, and branch locators — all handled in Bangla.

The Business Model

DeshiBot runs on a monthly subscription based on conversation volume. Onboarding includes training on company-specific data, integration setup, and support. The goal isn't to replace human customer service — it's to automate the repetitive 70-80% of queries so human agents can focus on the cases that genuinely need them. Contact us at deshibot.com to get started.

In 2024, Bangladesh was the world's second-largest freelancer market by volume. Its software developers were building applications used by millions of people globally. Yet almost none of the AI products those developers used daily — the image generators, the language models, the automation tools — were built in Bangladesh.

That gap is what started Everest Information Technology.

The Frustration

We'd been working in tech in Dhaka for years. We watched the Bangladeshi software industry grow into a genuine force — talented engineers, competitive output, a thriving startup ecosystem. But we kept seeing the same pattern: Bangladeshi companies were building on top of foreign AI infrastructure. Wrapping foreign APIs. Reselling foreign tools.

There's nothing wrong with that as a business. But it means that as AI becomes the defining technology of this decade, Bangladesh stays as a consumer of AI rather than a builder of it. The IP, the data, the model weights — all controlled from abroad. We thought we could change that.

Why a Holding Company

The decision to structure as a holding company from day one was deliberate. The most expensive parts of building an AI company in 2025 are GPU compute, AI engineering talent, DevOps infrastructure, and payment integrations. If you build one product, you bear all those costs alone. If you build a holding company with shared infrastructure, the second product costs dramatically less to launch than the first.

DeshiBot was a new product on existing infrastructure — not a company that had to rebuild everything from scratch. That's the holding company advantage.

What We've Learned

The holding company model works — but only if you're genuinely sharing infrastructure, not just sharing a brand name. The key is designing systems modular enough to serve multiple products without becoming a monolith that slows everything down.

We've also learned that building in Bangladesh means accepting constraints that don't exist in Silicon Valley — different payment infrastructure, different cloud latency, a developing talent pipeline. But these constraints have pushed us to solve problems that will matter enormously as the rest of the emerging world builds AI companies.

What Comes Next

We're evaluating a third portfolio company, expanding our Bangla NLP capabilities, and actively looking for engineers, researchers, and operators who want to build Bangladesh's technology future — not just outsource it. If that's you, reach out at info@everinfotech.com.