Talyn
Regular Member
- Oct 22, 2014
- 216
- 74
Hey all,
After making good money with fiction books on KDP for years, it's time for me to embark on a new journey. I got tired of dealing with ghostwriters and the fact that the life cycle of non-fiction books is getting shorter and shorter.
I have no proper CS education, even if I have a few agile certs, but learned "on the field" for years, I'm not even halfway decent in coding, but I'm still learning every day.
At the beginning of this year, I started playing with Chat GPT, which was getting a lot of hype. Since I can code a bit in Python, I started playing around with the unofficial API at that time and then moved on to the official Davinci and recently the GPT3.5 API.
As of today, I have created three sites:
After making good money with fiction books on KDP for years, it's time for me to embark on a new journey. I got tired of dealing with ghostwriters and the fact that the life cycle of non-fiction books is getting shorter and shorter.
I have no proper CS education, even if I have a few agile certs, but learned "on the field" for years, I'm not even halfway decent in coding, but I'm still learning every day.
At the beginning of this year, I started playing with Chat GPT, which was getting a lot of hype. Since I can code a bit in Python, I started playing around with the unofficial API at that time and then moved on to the official Davinci and recently the GPT3.5 API.
As of today, I have created three sites:
- The first website:
I used the official Davinci-003 API and unofficial/wrapper ChatGPT (which was a pain in the ass to deal with because of incomplete answers). The algorithm was really basic, just a list of keywords and ideas to use for each keyword in a CSV file. The niche is kind of related to finance and it's not getting a lot of traffic (and the search volume in google keywords is non existent).
- The second website focuses on as few uncommon niches:
It's a pure PAA website, with 700-800 words articles. I switched to the GPT-3.5 turbo and I introduced new few functionalities, like pushing articles to WordPress automatically, basic article outlining process (pictures, videos, etc.)
So far, the website has 5k pages indexed and is getting some decent traffic.
This website also pushed me to work on a solution for indexing pages faster. I implemented the functionality to send the URL to GSC with the Google API, and so far, it works well.
- The third site is a PAA/education site in a not too narrow niche.
This time I focused on my PAA scrapper, which feeds a an input CSV and then creates engaging but basic outlines before prompting/generating articles. It was a great introduction to Re3 and how to deal with numerous prompts to generate a consistent article (without getting millions of introductions/conclusions in the same article). This website is getting decent traffic with 600 articles ranging from 1500 to 12000 words long. For now, I don't plan to generate more articles as I hit a wall with content generation and am struggling to get a good algorithm for Re3. I'm still not happy with my outputs, and I'd like to have GPT3.5 generate more "storytelling" articles. I'm actively working on it.
The whole site got generate within 72h and got indexed within 48h so my way to push the articles to google seems to be working
- The fourth website is in a highly competitive niche, but with readers who have a tendency to binge shop.
In order to rank better, I'm focusing on keyword volume and interlinking. Here's what I'm working on:
- How to validate keywords (cosine analysis)
- Clustering keywords (maybe K-Means? DBSCAN? Hierarchical clustering? with or without using embeddings?)
- Interlinking (cosine and embeddings?)
- How to write titles on thumbnails + quality control (pillow)
- Code architecture, even if I commented it, it's a mess, and as I added more functionalities, I don't want to turn it into something impossible to maintain.
- Improving the content quality with Re3. I'm considering hiring a tutor/teacher to learn faster.
- Improving quality control before pushing the articles.
- Streamlining to be able to feed a few keypoints/data and let it run alone with some gate controls.
- SQL: several CSVs with several thousand lines is a pain to deal with
- Performance analysis. So far, I track the time it took for each article and the word count.
- UI, command line is cool, but not fun, I'd like to have a nice dashboard.
- Multithreading with concurrent.futures / ThreadPoolExecutor as the bottleneck is the API and not the processing power if I don't use any NLP/LLM in local like alpa.