Chapi Dev Talks
Open in Telegram
My name is Chapi and I am a Developer. I post my thoughts about tech here. Message to our assistant to give U Feedback: @sophiservebot Join https://t.me/chapidevtalks_group
Show more📈 Analytical overview of Telegram channel Chapi Dev Talks
Channel Chapi Dev Talks (@chapidevtalks) in the English language segment is an active participant. Currently, the community unites 10 275 subscribers, ranking 11 450 in the Technologies & Applications category and 3 231 in the Ethiopia region.
📊 Audience metrics and dynamics
Since its creation on невідомо, the project has demonstrated rapid growth, gathering an audience of 10 275 subscribers.
According to the latest data from 20 September, 2026, the channel demonstrates stable activity. Although there has been a change in the number of participants by 310 over the last 30 days and by 2 over the last 24 hours, overall reach remains high.
- Verification status: Not verified
- Engagement rate (ER): The average audience engagement rate is 33.46%. Within the first 24 hours after publication, content typically collects 17.68% reactions from the total number of subscribers.
- Post reach: On average, each post receives 3 438 views. Within the first day, a publication typically gains 1 817 views.
- Reactions and interaction: The audience actively supports content: the average number of reactions per post is 28.
- Thematic interests: Content is focused on key topics such as developer, database, dataset, addis, dataset.et.
📝 Description and content policy
The author describes the resource as a platform for expressing subjective opinions:
“My name is Chapi and I am a Developer.
I post my thoughts about tech here.
Message to our assistant to give U Feedback: @sophiservebot
Join https://t.me/chapidevtalks_group”
Thanks to the high frequency of updates (latest data received on 21 September, 2026), the channel maintains relevance and a high level of publication reach. Analytics show that the audience actively interacts with content, making it an important point of influence in the Technologies & Applications category.
10 275
Subscribers
+224 hours
+387 days
+31030 days
Data loading in progress...
Similar Channels
Tags Cloud
Incoming and Outgoing Mentions
---
---
---
---
---
---
Attracting Subscribers
September '26
September '26
+261
in 54 channels
August '26
+329
in 32 channels
Get PRO
July '26
+495
in 25 channels
Get PRO
June '26
+311
in 8 channels
Get PRO
May '26
+349
in 22 channels
Get PRO
April '26
+370
in 20 channels
Get PRO
March '26
+266
in 91 channels
Get PRO
February '26
+204
in 35 channels
Get PRO
January '26
+477
in 55 channels
Get PRO
December '25
+342
in 21 channels
Get PRO
November '25
+323
in 41 channels
Get PRO
October '25
+250
in 16 channels
Get PRO
September '25
+261
in 9 channels
Get PRO
August '25
+399
in 25 channels
Get PRO
July '25
+494
in 39 channels
Get PRO
June '25
+161
in 6 channels
Get PRO
May '25
+131
in 8 channels
Get PRO
April '25
+131
in 8 channels
Get PRO
March '25
+112
in 10 channels
Get PRO
February '25
+304
in 46 channels
Get PRO
January '25
+336
in 47 channels
Get PRO
December '24
+329
in 17 channels
Get PRO
November '24
+354
in 22 channels
Get PRO
October '24
+388
in 28 channels
Get PRO
September '24
+514
in 17 channels
Get PRO
August '24
+455
in 15 channels
Get PRO
July '24
+328
in 19 channels
Get PRO
June '24
+263
in 14 channels
Get PRO
May '24
+340
in 10 channels
Get PRO
April '24
+422
in 13 channels
Get PRO
March '24
+257
in 5 channels
Get PRO
February '24
+225
in 2 channels
Get PRO
January '24
+681
in 10 channels
Get PRO
December '23
+670
in 6 channels
Get PRO
November '23
+279
in 7 channels
Get PRO
October '23
+53
in 0 channels
Get PRO
September '23
+58
in 0 channels
Get PRO
August '23
+24
in 0 channels
Get PRO
July '23
+85
in 0 channels
Get PRO
June '23
+197
in 0 channels
Get PRO
May '23
+152
in 0 channels
Get PRO
April '23
+1 350
in 0 channels
| Date | Subscriber Growth | Mentions | Channels | |
| 21 September | 0 | |||
| 20 September | +5 | |||
| 19 September | +5 | |||
| 18 September | +1 | |||
| 17 September | +6 | |||
| 16 September | +2 | |||
| 15 September | +3 | |||
| 14 September | +24 | |||
| 13 September | +14 | |||
| 12 September | +6 | |||
| 11 September | +13 | |||
| 10 September | +1 | |||
| 09 September | +4 | |||
| 08 September | 0 | |||
| 07 September | +18 | |||
| 06 September | +6 | |||
| 05 September | +1 | |||
| 04 September | +20 | |||
| 03 September | +20 | |||
| 02 September | +105 | |||
| 01 September | +7 |
Channel Posts
Good Morning fellas
Have a great week ahead!
Try Hohe ASR: https://dataset.et/ai
What a great way to start the day
| 2 | Good Night enough for today!
https://t.me/chapidevtalks/3038 | 493 |
| 3 | BTW this is hosted on SPOT CPU
4 vCPU (2 physical AMD EPYC cores at 3.6 GHz), 8 GB RAM, up to 12.5 Gbit networking, 30 GB of gp3 disk!
CTC is amazing for this kind of stuff and the Speed median 1.05 s · 9 in 10 under 3.77 s
super fast and i hope this will be a good addition to the open source model | 446 |
| 4 | support us by sharing the message
https://x.com/chapimenge3/status/2101789761031557562?s=20
https://lnkd.in/p/epe26bNi | 578 |
| 5 | Love to get feedback let us know | 549 |
| 6 | Hohe ASR is out. Open Amharic speech to text.
Start with the speed, because it is the part people do not expect. On four ordinary CPU cores, a five second voice note is transcribed in 0.39 seconds. A thirty second clip takes 7.75 seconds. An hour of audio goes through in under ten minutes, on a machine that costs about nine cents an hour to rent. No GPU anywhere in that sentence.
It is a CTC model, one single pass over the audio instead of generating word by word, which makes it roughly five times faster than a Whisper of similar accuracy. It also means it does not invent sentences out of silence.
Accuracy: 16.1% word error and 5.4% character error on speakers it has never heard. 880 hours of training audio, including five regional dialects and a third of it pushed through a phone line on purpose.
What it is bad at, plainly: two people talking at once, around 47% word error. Numbers come out as words. No punctuation.
Try it now, nothing to install: @dataset_ai_bot
Weights, open under CC BY 4.0: huggingface.co/snapwre/hohe-asr-amharic
Send us what it gets wrong. | 816 |
| 7 | New Open source model releasing tonight | 565 |
| 8 |  software guy on seyfu 🔥 | 877 |
| 9 |  new home | 1 344 |
| 10 |  Sad to see let it go!
i will be back soon untill then adios! | 1 362 |
| 11 |  The model doesn't do well on interpreted conversations we are gathering more data and the more we train on that part the more we make it accurate!
It is also very fast model to actually get the transcript of 1,548 audio clips (short audio) it took 58 seconds
To put it on comparison the best open source model by Addis AI shook model tool around 298 seconds!
We got encouraged and motivated by Addis AI on this process a lot. Honestly big W for them!
Thank you | 1 756 |
| 12 | So the past couple of weeks quietly I was training a model that can have a CER (character error rate) and WER (word error rate) for Amharic speech to text models and pretty much there is so many out there but not quite good one so far.
So our new open source model that I will be releasing early next week is achieving almost less than 20% of WER and about 6% CER on speech it has never heard, from speakers that were never in the training data.
To put that next to what is out there: on the same test set, the best open Amharic model we could find scores 27.7% WER. Ours scores 18.7%. On FLEURS, the public benchmark, ours is at 17.8%. It also runs about five times faster, because it reads the audio in one pass instead of guessing word by word, and it does not hallucinate text when nobody is speaking.
One thing I want to be honest about: this is for clear speech. On real conversation, people interrupting each other, street noise, someone talking to a phone camera, it is still around 50% WER. That is the next problem, and it is a data problem, not a model problem.
Which is why the release includes more than weights. Everything it learned from, how it was measured, and what it still gets wrong.
If you have Amharic audio, or you want to test it on your own data, reply or DM me. Especially if you have recordings of people actually talking, not reading.
Free, open, and built here. | 1 821 |
| 13 | Juma Mubarak | 2 199 |
| 14 | Well and Alive here 😆 | 2 722 |
| 15 |  🛡 The CTF may be a competition, but the learning doesn’t stop at the final flag.
Ethiopian CyberShield 2026 CTF participants will join an exclusive panel discussion on “Ethiopia’s Cyber Future: Cloud Resilience, AI Security, and the Next Generation of Defenders.”
4 experts. 1 moderator. One conversation shaping tomorrow’s defenders. 🚀
Our new flexible, self-paced program is now live — discover your path with ALX at https://tinyurl.com/w575nsha
#ALXEthiopia #ALXAfrica #LifeAtALX #DoHardThings #CTF2026 | 2 811 |
| 16 | So guys show some love like, comment and repost 🙏.
https://vt.tiktok.com/ZSq4TGYQ1/
Dataset.et | 3 707 |
| 17 | Damnnnn | 3 530 |
| 18 | What the heck my claude Max usage limit hit without even doing a proper result?
Is Codex Max give more usage limit? | 3 652 |
| 19 | Good Morning y'all
Have a great working day ahead.
This year first official work day, let this year be successful year for all of us. | 4 981 |
| 20 | @Robi_makes_stuff odit users are honestly rich AF 😂😂😂
Standing tells me all! | 3 874 |
