Kurdish ASR, Life After AI, and New Opportunities

Participants: Sirwan, Omid, Arash, Saman

We started with a short chat about October’s unpredictable weather 😄, job opportunities, and Omid’s progress over the past two weeks. Sirwan then suggested three regular meeting topics: KurdSoftware projects, AI news, and life after AI. We also discussed recent news about Tesla, Elon Musk, and flying cars. Everyone was amazed by how quickly technology is advancing.

KurdSoftware ASR Project

Sirwan created asr.kurdsoftware.com and a Telegram bot to collect Kurdish speech recordings. The first model training test gave poor results. We reviewed some examples from the Sina Ahmadi dataset and found problems with spelling, transcription quality, and language consistency. Kurdish has many regional differences, which makes this more difficult.

We decided not to use the Sina Ahmadi dataset as our main data source. Instead, we will collect our own recordings through the KurdSoftware Telegram community, with help from Arash. Arash will also share a Kurdish spell checker.

We discussed support for different accents, including Kamyaran and Kalhor. Omid can help us collect samples from other regions. The plan is to first create a reliable baseline dataset that we can use as our single source of truth. We can then add other accents and use tags to identify them. We also agreed to keep punctuation for now because we can remove it later during preprocessing.

Life After AI

A large part of the meeting focused on how we can prepare for life after AI. Most of us live outside our home country, where living costs, mortgages, and other financial responsibilities are high. We discussed the risk of depending only on software engineering as our main source of income.

Sirwan explained that many people still evaluate AI by looking at what one model can do in a single session. However, this does not consider systems with multiple agents that work together. A swarm of agents can divide tasks, review results, and complete complex projects.

• Redesigning source control at Anthropic \ Anthropic
• Crafting Apps: open-source creative tools — ArtCraft

We also discussed the idea that AI engineers will remain necessary to manage agents, build pipelines, and evaluate their results. Sirwan suggested that this work could also become automated. Tools such as Grok and OpenAI Dots show how AI systems are moving towards more independent operation. In the future, companies could use large groups of agents to manage tasks that currently require human coordination.

This raised an important question: If AI can build software and manage other AI agents, what skills should we develop to remain financially independent?

Farming and Alternative Careers

Arash prefers practical projects and suggested electronic equipment repair and farming as possible alternatives. He also explained that AI adoption in Iran is still limited compared with other countries. He has considered buying farmland in Iran but is concerned that it could be difficult to sell later.

Omid shared his insights with farming in Sweden. He discussed company registration, government support, hiring workers, and business taxes, including Sweden’s standard VAT rate of 25%. He also talked about wheat, maize, and rye. According to Omid, farmers working a wheat field opposite his house need only about three days of active work because they have the necessary machines. Saman also expressed strong interest in farming.

YouTube and New Opportunities

We discussed YouTube as a possible source of income, especially because we already have experience with AI tools. Creating videos is now much easier, and AI can help with almost every part of the process.

Sirwan shared his recent experiment with two short videos about the UK. The first video received a few thousand views in less than one day, which was an encouraging result. He created the videos by connecting Claude Code to Higgsfield through MCP and using AI video generation models.

We discussed how these tools make content creation more accessible and how we could use them to create Kurdish content, reach a larger audience, and potentially earn money through YouTube. Monetisation would still depend on audience growth and YouTube’s eligibility requirements.

Final Thoughts and Next Steps

We agreed that building products and experimenting with new ideas is easier than ever. Instead of only discussing how AI could change our jobs, we should use these tools to create new opportunities, especially for the Kurdish community.

• Sirwan: Continue improving the ASR website, Telegram bot, and data collection process.
• Arash: Share the Kurdish spell checker and help collect reliable recordings.
• Omid: Help explore recordings from additional Kurdish accents.
• Team: Build a reliable baseline dataset before adding accent tags and more regional variations.
• Team: Explore practical AI projects, Kurdish content creation, YouTube, and alternative sources of income.

The main conclusion was simple: We should continue to learn, experiment, and build while preparing for a future in which AI can do much of today’s software engineering work.