Adhish Chitranshi

Trying to build something deterministic in the probabilistic world
Building products & systems | distributed systems & scalable design

[Adhish Chitranshi @020Adhish]

@system_monarch I don't think 50M API calls would be made, Zomato calls the firebase once, and then the firebase announces it to the topic (the topic contains the devices registered for that notification )

[Adhish Chitranshi @020Adhish]

@system_monarch "Indexes make reads faster."
True.
But they also make the database do extra work whenever data changes.
https://t.co/wgbfAc1oIS

[Adhish Chitranshi @020Adhish]

Learning database internals 📚

Imagine tracking a user's profile.

User changes:
📍 Location
📷 Profile picture
📝 Bio
100 times.
A log-structured database doesn't update the old record—it keeps appending new versions.
Later:
🔹 Compaction removes outdated versions
🔹 Merging combines cleaned segments
🔹 Hash indexes point directly to the latest profile data

Simple idea. Powerful architecture.

Now I understand why systems like Bitcask are so fast 🚀

[Adhish Chitranshi @020Adhish]

Learning database internals 📚

Imagine tracking a user's profile.

User changes:
📍 Location
📷 Profile picture
📝 Bio
100 times.
A log-structured database doesn't update the old record—it keeps appending new versions.
Later:
🔹 Compaction removes outdated versions
🔹 Merging combines cleaned segments
🔹 Hash indexes point directly to the latest profile data

Simple idea. Powerful architecture.

Now I understand why systems like Bitcask are so fast 🚀

[Adhish Chitranshi @020Adhish]

Ever wondered WHY SQL databases became dominant? 🤔

The answer lies in how database models evolved over time 👇

1. Hierarchical Model

Best for one-to-many relationships
Data stored like a tree structure
Similar to how modern JSON documents look today
But handling many-to-many relationships became difficult

2. Network Model

A record could have multiple parents

Supported:
many-to-one
many-to-many relationships
Every record acted like a linked node

But there was a catch ⚠️
Developers had to manually manage the traversal paths (“access paths”) to fetch data.
Complex systems = painful queries.

3. Relational Model

This changed everything.
Data organized into tables/tuples

Relationships handled using foreign keys & joins

Introduced the game changer → Query Optimizer

Instead of developers manually deciding how to fetch data,
the database automatically chooses:
✅ best indexes
✅ optimal execution order
✅ fastest access path

That abstraction is one of the biggest reasons SQL databases scaled in adoption. 🚀

Understanding database evolution helps you understand why modern systems are designed the way they are.
#SQL #NoSQL #Databases #SystemDesign #Backend #SoftwareEngineering #DataEngineering #Programming #TechTwitter

[Adhish Chitranshi @020Adhish]

@system_monarch Thanks for such a wonderful share.
I am even trying to understand the distributed systems
Written a small article related to caching -

https://t.co/gI0EVDFHBx

Would really appreciate some insights.

[Adhish Chitranshi @020Adhish]

https://t.co/R0DeT52iEj

[Adhish Chitranshi @020Adhish]

Spent this weekend diving into research papers.
Just published my first article on caching, summarizing “Scaling Memcache at Facebook.��

https://t.co/gI0EVDGfr5

Curious to know—have you explored better approaches? Let’s discuss

[Adhish Chitranshi @020Adhish]

https://t.co/R0DeT52iEj

[Adhish Chitranshi @020Adhish]

https://t.co/R0DeT52iEj

[Adhish Chitranshi @020Adhish]

@abhijitwt They’re volatile — everything lives in memory, so a restart wipes it out
No real persistence guarantees, unlike databases that use things like WAL(write - ahead logs)
RAM is expensive compared to disk, so scaling can get costly

[Adhish Chitranshi @020Adhish]

@Ai_Vaidehi Thanks for this @Ai_Vaidehi

I would really appreciate it if I could get some insights on this

https://t.co/uG4kuyTSrv

Thank you

[Adhish Chitranshi @020Adhish]

Getting image generations to match a reference layout is still tricky—models tend to hallucinate placements. Any tips on reducing that and getting outputs that stick closely to the reference + prompt?

[Adhish Chitranshi @020Adhish]

Getting image generations to match a reference layout is still tricky—models tend to hallucinate placements. Any tips on reducing that and getting outputs that stick closely to the reference + prompt?

[Adhish Chitranshi @020Adhish]

System Design fundamentals every engineer should know:
When building scalable systems, the first question is: single server or distributed?
Start with a single server — it's simpler, easier to iterate on, and perfectly fine for early-stage products and predictable loads. But once you hit 10x growth, things change fast.
At scale, you need:
→ Horizontal scaling + sharding for data
→ Load balancers to distribute traffic
→ Database replication (master for writes, replicas for reads)
→ Cache layers (read-through strategy, LRU eviction)
→ CDNs for static content delivery
→ Stateless architecture so any server can handle any request
→ Message queues to decouple services asynchronously
→ GeoDNS + multiple data centers for availability & latency

Choosing the right database matters too:
• Relational DB → complex joins, many-to-many relationships
• Document DB → self-contained "tree" data (one-to-many)
• Graph DB → highly interconnected, "anything relates to everything"

And don't forget observability:
Logging + metrics (host, aggregated, business KPIs) + CI automation = a system you can actually trust.
The hardest part isn't building it. It's knowing when and what to scale. 🚀
#SystemDesign #SoftwareEngineering #ScalableSystems #BackendEngineering #TechLearning

[Adhish Chitranshi @020Adhish]

@freeCodeCamp @manishmshiva cool

[Adhish Chitranshi @020Adhish]

@EazeMyEvent Lets go !!

[Eaze My Event @EazeMyEvent]

We’ve launched our new website with a fresh new theme.
Please check it out: https://t.co/vZP0ZsTebg

Would love to hear your feedback. Thanks Everyone for all the support! 🙌

[Adhish Chitranshi @020Adhish]

State Management Principle, which I just encountered

[Adhish Chitranshi @020Adhish]

This is when dot product stopped being “math”.
It became:
normalization
intent extraction
constraint logic

A way to ask:
“How much of this belongs in that direction?”

[Adhish Chitranshi @020Adhish]

A math formula I memorized years ago finally turned into code intuition.
For the longest time, dot product was just a formula I used.
A⋅B = ∣A∣∣B∣cos⁡θ
I could calculate it.
I couldn’t feel it.

[Adhish Chitranshi @020Adhish]

And now t tells you everything:

t = 0 → start

t = 1 → end

0 < t < 1 → somewhere in between

< 0 or > 1 → outside

All from one line.