Samri-A
Numbered days of code, LeetCode dailies and Medium posts on AI ideas, all logged in public.
Hey, I’m Samri 🤗. in one word "someone curious" This channel is my log book.
Samri-A has been posting since September 2024, mostly in English. It counts its days out loud, from Day 12 binary search drills to Day 67 dynamic programming, and once invited readers into a month of LeetCode daily challenges. Between logs it publishes Medium posts on building a neural network from scratch with NumPy, on embeddings and on RAG.
- Its longest posting streak ran 34 consecutive weeks.
- The channel's favorite word is "day", which suits a log full of numbered days.
- 577 posts have been collected from the channel so far.
The journey.
- Jul – Dec 2025
Notebooks to NumPy
The channel finds its footing by sharing ML tools, praising MLflow for tracking experiments and pointing to a free data science course at wqu.edu. By December it has published its own Medium posts on a neural network built with only NumPy and on embeddings, inviting corrections.
Sources#40#188#199#214 - Mar – Jun 2026
One day at a time
A 100-day self-development challenge gets logged one numbered day at a time: binary search, half of Grokking Algorithms, graphs and recursion. In June the channel invites readers into a month of LeetCode daily challenges, and by Day 67 it is on dynamic programming and reading up on Fenwick trees.
Sources#308#319#373#462 - Jul 2026
Project days
The numbered days turn into project days: when the first approach does not work, a mini version with a new approach is built, then scaled up and given a waitlist. Along the way the channel notes that the hardest part of building an AI project is a strong knowledge base, and shares a sneak peek of a portfolio rebuild.
Sources#542#581#585#586 - Aug – Oct 2026
Go, Kubernetes and a pause
After a break, the channel returns with its first open source contribution, all three lines of it. It publishes a fourth blog on RAG, logs learning Go and system design, tries Kubernetes for the first time and joins a competition to start modeling. In October something with an over-engineered backend finally goes live, and the channel announces it will be inactive on Telegram for six months.
Sources#596#597#619#622
Start here.
- 01
Everything the channel knows about embeddings, written up on Medium with corrections welcome.
I’ve written down all the little things I know about embeddings. If you have any suggestions or corrections, feel free to comment. https://medium.com/@samrawit...
Dec 31, 20251.1K views - 02
A neural network built from scratch with only NumPy, explained in a Medium post.
https://medium.com/@samrawitasfaw8/building-neural-network-from-scratch-using-numpy-only-e825fcaac454 It’s been two weeks since I published this, but I wanted...
Dec 1, 2025660 views - 03
The fourth blog, on RAG, published after months as a draft.
I published my 4th blog. This is not as deep as the previous one, but yes, after keeping it as a draft for months, it is finally published... https://medium.co...
Sep 10, 2026250 views - 04
The kickoff for a month of LeetCode daily challenges that anyone can join.
If you can't be consistent on LeetCode, today is the day to start. It's June 1st, let's tackle the Daily Challenge consistently for a month. Today's problem is...
Jun 1, 2026502 views - 05
A typical numbered log: scaling a mini architecture, building a waitlist, and the LeetCode daily.
Day 80 : yesterday Scaled the mini architecture Build the waitlist Researching for partnership thing Day 81: today Still on that project ( python repo only w...
Jul 16, 2026439 views
In their own words.
The hardest part of building an AI project is building a strong knowledge base
I sometimes feel like I m not improving but if I stop I won’t improve
Fresh from the channel.
I will be inactive on Telegram for the next 6 months ( and I m fine it is for the best ) . If I find anything valuable, I will come back to post it. Wherever things start lining up or going well, there will be something in your way to ruin...
the second one pending
after procrastinating , over enginnering the backend ( well i wasted my azure credit with 3 vm😐) it now live
Go with the flow The Advice Changed My Life My sister wrote this beautiful substack
"Not ready" is biggest self scam 🙃
after using pretrained model