Icon
Back to Blog

Intro to solopreneurship: Indie Making of Llanai

Introducing a series on the making of Llanai, as a solopreneur and the motivation behind it.

Posted by

A screenshot of the report by the S&P500 on the number of tech workers needed to generate $1M revenue
A lot of value can be generated by a single person [1]. Source: Bank of America

Ari here. A solopreneur in the intersection of Web Development and Artificial Intelligence.

In this series, I will detail the making of Llanai, my language learning chatbot for improving one's speaking skills from the lenses of:

  • User & Market Research - Is there a need?
  • Product Design - Making pleasant experiences in generative AI
  • Prompt Engineering - Replicating a good language teacher
  • Solution Architecture - Scalable tech stack design
  • Challenges and lessons learned - What would be done differently?
  • Funding your vision - Bootstrapping the proper way

My journey

For introduction's sake, I grew up in Rhodes, Greece, but my entire adulthood was in the USA. Here's a brief timeline:


  • Rhodes Island, Greece
    Blessed to have been raised in the historic Rhodes.
    Home to the fallen Colossus of Rhodes and the Old Town built by the Knights of Saint John.
  • Rhode Island, USA
    Life has its ways; I moved from Rhodes Island to Rhode Island to study at Brown University.
    Completed Sc.B. in Chemical Engineering.

    The international community was the best part about college.
  • SF, California
    Took a road trip from Rhode Island to California.
    Ended up becoming an 11 year stay, which started by teaching STEM.
  • Chemical Water Treatment
    Hired by Garratt-Callahan for HVAC (heating, ventilation and cooling) system maintenance.
    These systems are a cornerstone of modern building design.

    Chemical maintenance is needed, because water is corrosive and a pathogen carrier!
  • Impossible Foods
    On August 2016, I joined R&D to create plant/microbial-based ingredients
    for animal products substitutes

    I was part of the team that scaled heme manufacturing
    from 100L->100,000L scale.

    Spent countless nights running experiments & ops across the Bay and Mexico.

    Also, spearheaded workflow automation by co-creating a data pipeline
    using Gsheets, GCP, and Streamlit.
    This was my first major project in web development.

    Fun fact: All organisms have hemoglobin-like proteins, as they are vital for oxygen transport.
  • AI Engineering
    Let go by IF, departed California and founded my own practice.

In February of 2023, time was abundant and the internet was running haywire with ChatGPT-related news. So, I eagerly embarked on a new (solo) chapter.


Decided to use web development as my sandbox to build in public, which is quite common(hint: 161k members in Sideproject subreddit). With my eyes set to the future of how we work as humans, I chose to focus on the nascent discipline of AI Engineering. Essentially, it is the role that entails picking the right models (gpt-4o, opus, llama 3, etc.) and tools (databases, memory, wrappers, etc) to develop an AI application. There's a primer by the super sharp @swyx that I recommend.


Largely a reflection of my state of mind, my first project was a trip planner on Streamlit. Although happy with the learnings, I shelved it and did not pursue commercialization. Below you can see a typical island hopping adventure in the Dodecanesean (Twelve Islands) region of Greece, with my native Rhodes.


Screenshot of the Travel Recommendation app on Streamlit
A Streamlit app I built to plan my travels in the summer of 2023.

The Birth of Llanai

Learning languages is intrinsic to my identity. My mother tongues are both (American) English and Greek. I learned German in middle school, because Rhodes has lots of Bavarian visitors. I picked up Spanish and Japanese, while in California. Thus, the trend of learning language throughout my life.However, after years of studies, I realized that sounding natural, while being grammatically correct are my biggest hurdles out of the pre-intermediate zone.


So on July 1st, 2023, with experience building LLM pipelines and after months of chatting with ChatGPT in Spanish (and a bit of Japanese), I pivoted to build Llanai, a language learning partner, available any time, on WhatsApp.

Why Llanai Over ChatGPT?

User interfaces define user behavior. ChatGPT innovated on conversational dialog flows by catering to a primal need of every human - information retrieval. While you can ask anything, you can easily get lost by the sea of threads you created. Also, there's a considerable likelihood of falsehood in factual questions based on historical events and especially so in mathematical reasoning.

Llanai, on the other hand, is purely a language teacher. It is focused on conversing in high resource languages, such as Spanish, German, French, and English. You can read more about the benchmarks that OpenAI and other Frontier model makers (Meta, Anthropic, Cohere, Google, etc) use here. These assessments are not a panacea, although they may pass the Bar exam, they fail simple riddles.

Llanai is inherently superintelligent, however its clean interfaces for language exchange and linguistic performance tracking differentiate it from ChatGPT.
• Llanai's WhatsApp interface is a dedicated chat window, simulating a 1-1 lesson.
• Llanai periodically revisits your linguistic performance and makes recommendations to meet your learning goals.
• Also, you can see your fluency changing over time! I don't believe in meaningless points like Duolingo does.


Screenshot of Llanai's grammar tracking feature
Llanai offers a dedicated portal to track your linguistic performance and progress.

The Power of Prompt Engineering

Later in the series, I will delve into how Llanai was made, but the key innovation here lies in the LLM prompts.

Prompts steer the reasoning capabilities of an LLM. Our ability to articulate is not just a life skill; it is the de facto coding language of the future, as famously said by NVIDIA CEO Jensen Huang.

Moreover, we are interacting with a new technology that has surprising emergent behaviors and displays intelligent output. It is also 'unpredictable', due to its stochastic nature. This was a problem in the designing of Llanai, and can occassionaly still be. I solved one of Llanai's common deviations by using emotional stimuli (some may call it a bribe). Using positive language has statistically significant effect on OpenAI's models.

Conclusion

This is a major milestone in my journey to sustainable solopreneurship. By leveraging the power of LLMs and prompt engineering, Llanai offers a tailored experience for language learners, with features like progress tracking, personalized recommendations, and pronunciation assistance.

As I continue this series, I will dive deeper into the various aspects of building Llanai, from user research and product design to the technical challenges and lessons learned along the way.


[1] I do not mean to produce sensational material and overlook fundamental economics. The dollar had an average inflation rate of 2.54% per year between 1990 and today, producing a cumulative price increase of 128.90%. So $1M today was worth ~$437,000 in 1990. Still, it is pretty impressive that a single person entity can generate $1M in revenue in the tech sector.