Building Sodura AI: A Founder's Playbook

As a first-time founder, the hardest part is not building. Building has become easier. AI has collapsed software creation time, and a small team can now create products that previously needed a large company.

The harder question is: what should we become? For Sodura AI, the answer is becoming clearer. We are an AI product studio — not a company that builds random demos, but one that repeatedly identifies painful problems, builds AI-native products quickly, tests them with real users, and turns the winners into reliable businesses.

Our goal is not to have many half-working products. Our goal is a system where one product becomes revenue, another becomes strategic learning, and every experiment improves our shared AI platform.

---

Part 1: Company Direction

Huuli.tech is the flagship

Among all our products, Huuli.tech already has demand.

It has 50+ paid users. Legal professionals are using it. The pain is real. The market is clear. The product can go much deeper than simple legal Q&A.

So for the next stage, Huuli.tech must become the foundation of Sodura AI. My role as founder should reflect that — I should spend around 60% of my time on Huuli.tech until it becomes a reliable income source: first ₮10M/month, then ₮30M/month, and eventually much more.

Huuli should become the default AI workbench for Mongolian legal professionals: legal search, court decision analysis, contract review, legal memo drafting, claim drafting, citation-backed answers, and team workflows for law firms and institutions.

The question every week should be simple:

Did Huuli become more useful for real legal work?

If the answer is yes, the company is moving forward.

Sodura is a portfolio, but not chaos

Sodura AI can have multiple products, but each one needs an owner.

SoduraSpeak, the AI platform, and new business exploration are all valuable — but they cannot all depend on the founder's direct attention every day. Each product experiment should have a dedicated leader, almost like a mini-founder.

A product leader owns:

• the users
• the metric
• the roadmap
• the revenue target
• the weekly progress
• the decision to continue or stop

No owner means no product. No user demand means no long-term commitment. No revenue path means it stays an experiment. This is how Sodura avoids becoming a scattered company.

The AI platform is the hidden engine

Behind every product, we need a shared AI platform: RAG systems, OCR, legal embeddings, Mongolian language models, fine-tuning pipelines, evaluation tools, a citation engine, admin dashboards, and deployment infrastructure.

The platform's job is simple:

Make every new Sodura product faster, better, and more reliable.

Huuli improves the platform. SoduraSpeak improves the platform. Enterprise pilots improve the platform. New experiments improve the platform. Over time, this platform can become a business in its own right — local-language AI infrastructure for Mongolian companies and institutions. But first, it must make our own products stronger.

The founder's job must change

At the beginning, the founder builds everything. But the next stage is different.

My job is no longer just to build products. My job is to build the company system — Huuli growth, key customer conversations, sales, hiring, product quality, team ownership, company direction, and cash flow.

The founder should not be the bottleneck for every feature, every post, every bug, and every experiment. The company grows when other people start owning outcomes.

The next 12 months

One — Make the company story clear:

Sodura AI is an AI product studio building practical AI products for Mongolia's underserved markets. Huuli.tech is our flagship legal AI product.

Two — Make Huuli reliable:

Reach ₮10M/month recurring or repeatable revenue.

Three — Give each product a leader:

SoduraSpeak, AI platform, and new business exploration should each have a clear owner and a weekly metric.

Four — Build the internal studio system:

Every product must have a customer, a metric, a revenue model, and a continue/stop decision.

Five — Turn services and pilots into reusable platform components.

The company should not chase every exciting AI opportunity. It should test many things, but commit only to what shows real demand.

The end goal

The end goal is not just to build one app.

The end goal is to build Mongolia's strongest AI product studio — a company that can repeatedly turn local problems into useful AI products.

• Huuli.tech becomes the first reliable business.
• SoduraSpeak may become the education product.
• The AI platform becomes the internal engine.
• New experiments discover the next winner.

If we execute well, Sodura AI becomes known for one thing:

We make AI useful for real work in Mongolia.

That is the future: focused enough to grow, flexible enough to experiment, and disciplined enough to become a real company.

---

Part 2: Hiring the Best People

The best startup people are usually not actively applying for jobs. Here is where to find them — and how to tell if they are right.

Where to look

People already building things (highest signal) Look for people who have built a side project, created a Telegram bot, published an app, started a small business, run a Facebook page, organized a student club, or sold something online. Ask: "Show me something you've built." A student with a tiny project is often stronger than someone with a perfect GPA. University clubs and competitions

Programming clubs, debate clubs, entrepreneurship clubs, startup competitions, hackathons — the top 5% of students there are often much stronger than average job applicants.

Hackathons and startup events

People who spend weekends building things are your people: startup weekends, AI meetups, developer communities, university hackathons, tech conferences like Dev Summit. These people already have the mindset you are looking for.

Referrals from founders Ask every founder: "Who is the most impressive young person you've worked with?" One recommendation from a founder is often worth 100 CVs. In Mongolia's startup ecosystem, the network is still small enough that this works extremely well. Your own users (underrated)

For Huuli.tech, IELTS.sodura.ai, and future products — watch who sends thoughtful feedback, reports bugs, asks smart questions, and uses the product heavily. Many great startup hires begin as power users.

Where I would spend my recruiting time

One recruiting trick that works

Instead of asking for a CV, post a challenge:

"Build something useful with AI in 3 days and show us."

• Marketing candidate → create a campaign for SoduraSpeak
• Engineer → build a small AI tool
• Sales candidate → get 3 demo meetings booked

The people who respond to this are exactly who you want.

What we are looking for

Must have:

1. Proactivity — sees a problem and acts without being told
2. Ownership — treats company problems as their own problems
3. Learning speed — can become 2x better every few months
4. Resourcefulness — finds a way even without resources
5. Bias toward action — ships before they feel ready; builds a prototype today, talks to users tomorrow, improves next week

OK to have:

• Does not fit traditional corporate culture
• Impatient
• Makes mistakes
• No impressive credentials
• Messy and disorganized

Not OK to have:

• Ego that blocks learning
• Toxic behavior: drama, blaming others
• Low integrity

---

Part 3: Mongolian AI Should Be Voice First

Not because voice is trendy. Because Mongolia is a phone-call country.

When people need something, they do not write a careful message and wait. They pick up the phone. They call the doctor, the lawyer, the bank manager, the school, the supplier, the cousin who knows someone. A five-minute call replaces twenty messages.

That habit matters for AI.

Most AI products are still designed like writing tools: open a chat box, type a question, read the answer, edit the prompt, read again. That is natural for people who already live inside documents and keyboards. But in Mongolia, many everyday workflows are oral. People explain things out loud. They ask follow-up questions immediately. They interrupt. They clarify. They want the answer in the rhythm of a real conversation.

So if we want Mongolian AI to feel native — not imported — the first interface should be voice.

Voice is also the fastest way to impress. When a user calls an AI and it understands Mongolian, remembers the context, asks one useful follow-up, and gives a clear answer, the reaction is instant. It feels alive. It feels useful. It feels like the future arrived in the format people already trust.

This is especially true for professional AI:

• A lawyer should be able to ask about a law by voice.
• A business owner should be able to call and ask what changed in a regulation.
• A student should be able to practice speaking, not just reading.
• A customer should be able to solve a problem without navigating a menu or typing a paragraph.

The product lesson is simple: do not force Mongolian users into a foreign habit just because the model started as text. Start where the user already is. For Mongolia, that place is the call.

---

Part 4: AI Collapsed Build Time — Now What?

AI collapsed the time it takes to build software. That changes the startup strategy.

Before, choosing one vertical felt like a big commitment — months of building, launching, and learning. Now a small team can build a serious prototype in days or weeks. That means we should try several verticals at once instead of pretending we know the answer too early.

The obvious verticals are still obvious for a reason.

Legal tech, especially compliance, is attractive because the work is document-heavy, high-stakes, and painful. Language learning is strong because AI can give instant feedback, personalized practice, and voice-based coaching at a level old apps could not.

Some markets feel harder. Customer support is real, but crowded. Many companies are already selling AI agents, chatbots, helpdesk copilots, and call center automation. It may still be a good business, but it is not an empty field. Other areas deserve real experiments:

• HR and recruiting
• AI tender and procurement assistants
• Education beyond IELTS

The hardest question is not what to build first. The hardest question is when to stop.

AI makes building cheap, but it does not make focus cheap. Every vertical can look promising after the first demo. Every prototype can be improved one more time. Every weak signal can be interpreted as "maybe we just need another iteration."

So the discipline has to change. Build fast. Test fast. But decide in advance what proof you are looking for: paid pilots, repeated usage, urgent customer pain, or a workflow that clearly becomes better with AI.

If the signal is there, keep iterating. If the signal is not there, stop cleanly and move.

The opportunity is huge, but speed only matters if it creates learning. Otherwise, fast building becomes just a faster way to avoid choosing.