Rohit
@rohitlakh.bsky.social
Building digital products one bit at a time. Ex-Software Engineer | Growing System Design Newsletter (Read by 25k+ developers).
Newsletter: https://hw.glich.co/subscribe?utm_source=bluesky-profile
Newsletter: https://hw.glich.co/subscribe?utm_source=bluesky-profile
URI,URN,URL?
URL for web browsing and API endpoints;
URI for XML namespaces and RDF resources;
URN for persistent identifiers like ISBN, ISSN, and RFC numbers.
Each serves a distinct role in resource identification and access.
URL for web browsing and API endpoints;
URI for XML namespaces and RDF resources;
URN for persistent identifiers like ISBN, ISSN, and RFC numbers.
Each serves a distinct role in resource identification and access.
October 28, 2025 at 11:29 AM
URI,URN,URL?
URL for web browsing and API endpoints;
URI for XML namespaces and RDF resources;
URN for persistent identifiers like ISBN, ISSN, and RFC numbers.
Each serves a distinct role in resource identification and access.
URL for web browsing and API endpoints;
URI for XML namespaces and RDF resources;
URN for persistent identifiers like ISBN, ISSN, and RFC numbers.
Each serves a distinct role in resource identification and access.
An MVP isn’t the smallest product that works.
It’s the smallest product that sells.
It’s the smallest product that sells.
October 27, 2025 at 12:41 PM
An MVP isn’t the smallest product that works.
It’s the smallest product that sells.
It’s the smallest product that sells.
Debugging 101:
✅ Walk
✅ Shower
✅ Eat
✅ Nap
❌ Staring at the screen until your brain melts
✅ Walk
✅ Shower
✅ Eat
✅ Nap
❌ Staring at the screen until your brain melts
October 27, 2025 at 3:32 AM
Debugging 101:
✅ Walk
✅ Shower
✅ Eat
✅ Nap
❌ Staring at the screen until your brain melts
✅ Walk
✅ Shower
✅ Eat
✅ Nap
❌ Staring at the screen until your brain melts
A working product isn’t an MVP until it proves demand.
Minimum viable = minimum customers willing to pay.
Minimum viable = minimum customers willing to pay.
October 25, 2025 at 2:30 PM
A working product isn’t an MVP until it proves demand.
Minimum viable = minimum customers willing to pay.
Minimum viable = minimum customers willing to pay.
Password storage best practice: never store plain-text.
Instead, hash passwords with salt and store the hash.
On login, hash the input and compare.
Even if your database leaks, attackers face costly brute-force to reverse hashes.
Instead, hash passwords with salt and store the hash.
On login, hash the input and compare.
Even if your database leaks, attackers face costly brute-force to reverse hashes.
October 25, 2025 at 6:29 AM
Password storage best practice: never store plain-text.
Instead, hash passwords with salt and store the hash.
On login, hash the input and compare.
Even if your database leaks, attackers face costly brute-force to reverse hashes.
Instead, hash passwords with salt and store the hash.
On login, hash the input and compare.
Even if your database leaks, attackers face costly brute-force to reverse hashes.
Hardware vs software vs cloud load balancers:
Think expensive sports car vs versatile SUV vs rental car.
Hardware offers peak performance but costs more,
software provides flexibility,
cloud gives instant scalability.
Choose based on your needs and budget!
Think expensive sports car vs versatile SUV vs rental car.
Hardware offers peak performance but costs more,
software provides flexibility,
cloud gives instant scalability.
Choose based on your needs and budget!
October 24, 2025 at 12:48 PM
Hardware vs software vs cloud load balancers:
Think expensive sports car vs versatile SUV vs rental car.
Hardware offers peak performance but costs more,
software provides flexibility,
cloud gives instant scalability.
Choose based on your needs and budget!
Think expensive sports car vs versatile SUV vs rental car.
Hardware offers peak performance but costs more,
software provides flexibility,
cloud gives instant scalability.
Choose based on your needs and budget!
LinkedIn slashed datasets by 70% when moving from Teradata to Hadoop.
Why was this migration so tricky, and how did open-source tech unlock massive speedups up to 1000x?
The story behind the numbers is eye-opening...
Why was this migration so tricky, and how did open-source tech unlock massive speedups up to 1000x?
The story behind the numbers is eye-opening...
October 23, 2025 at 12:31 PM
LinkedIn slashed datasets by 70% when moving from Teradata to Hadoop.
Why was this migration so tricky, and how did open-source tech unlock massive speedups up to 1000x?
The story behind the numbers is eye-opening...
Why was this migration so tricky, and how did open-source tech unlock massive speedups up to 1000x?
The story behind the numbers is eye-opening...
You’ve got thousands of tables indexed, but half have no documentation.
An AI tries to help, but with missing info it guesses wrong most of the time.
How could you make finding the right table easier and actually useful for analysts?
An AI tries to help, but with missing info it guesses wrong most of the time.
How could you make finding the right table easier and actually useful for analysts?
October 23, 2025 at 3:31 AM
You’ve got thousands of tables indexed, but half have no documentation.
An AI tries to help, but with missing info it guesses wrong most of the time.
How could you make finding the right table easier and actually useful for analysts?
An AI tries to help, but with missing info it guesses wrong most of the time.
How could you make finding the right table easier and actually useful for analysts?
Project: Image Classification API
Knowledge: Python, TensorFlow/Keras, FastAPI
Topic: ML + API Development
Steps: Train CNN model on CIFAR-10 or custom dataset
Save model and write FastAPI endpoints for inference
Add image preprocessing (resize, normalize)
Knowledge: Python, TensorFlow/Keras, FastAPI
Topic: ML + API Development
Steps: Train CNN model on CIFAR-10 or custom dataset
Save model and write FastAPI endpoints for inference
Add image preprocessing (resize, normalize)
October 22, 2025 at 12:45 PM
Project: Image Classification API
Knowledge: Python, TensorFlow/Keras, FastAPI
Topic: ML + API Development
Steps: Train CNN model on CIFAR-10 or custom dataset
Save model and write FastAPI endpoints for inference
Add image preprocessing (resize, normalize)
Knowledge: Python, TensorFlow/Keras, FastAPI
Topic: ML + API Development
Steps: Train CNN model on CIFAR-10 or custom dataset
Save model and write FastAPI endpoints for inference
Add image preprocessing (resize, normalize)
After splitting payments into many independent services, engineers now face a new issue:
Just fetching a host’s earnings requires multiple API calls across services.
If you were leading, how would you simplify this data retrieval challenge?
Just fetching a host’s earnings requires multiple API calls across services.
If you were leading, how would you simplify this data retrieval challenge?
October 22, 2025 at 3:27 AM
After splitting payments into many independent services, engineers now face a new issue:
Just fetching a host’s earnings requires multiple API calls across services.
If you were leading, how would you simplify this data retrieval challenge?
Just fetching a host’s earnings requires multiple API calls across services.
If you were leading, how would you simplify this data retrieval challenge?
Don't use Cassandra for everything!
Overkill for small apps, bad for complex joins, not ideal for ACID-heavy transactions or read-heavy workloads.
It's a specialized tool - like using a race car for grocery shopping. Match the tool to the job for best results!
Overkill for small apps, bad for complex joins, not ideal for ACID-heavy transactions or read-heavy workloads.
It's a specialized tool - like using a race car for grocery shopping. Match the tool to the job for best results!
October 21, 2025 at 12:42 PM
Don't use Cassandra for everything!
Overkill for small apps, bad for complex joins, not ideal for ACID-heavy transactions or read-heavy workloads.
It's a specialized tool - like using a race car for grocery shopping. Match the tool to the job for best results!
Overkill for small apps, bad for complex joins, not ideal for ACID-heavy transactions or read-heavy workloads.
It's a specialized tool - like using a race car for grocery shopping. Match the tool to the job for best results!
An investor checks their trading app during a market surge.
Should the system always return the most accurate balance,
even if it means temporary unavailability, or serve a slightly stale view so they never see downtime?
Which risk would you accept?
Should the system always return the most accurate balance,
even if it means temporary unavailability, or serve a slightly stale view so they never see downtime?
Which risk would you accept?
October 21, 2025 at 11:29 AM
An investor checks their trading app during a market surge.
Should the system always return the most accurate balance,
even if it means temporary unavailability, or serve a slightly stale view so they never see downtime?
Which risk would you accept?
Should the system always return the most accurate balance,
even if it means temporary unavailability, or serve a slightly stale view so they never see downtime?
Which risk would you accept?
A company runs its backend on EC2.
But even with zero logins for weeks, the servers keep running and the bill never stops.
What cloud billing model would let them pay only when their code actually runs, instead of 24×7 idle server costs?
But even with zero logins for weeks, the servers keep running and the bill never stops.
What cloud billing model would let them pay only when their code actually runs, instead of 24×7 idle server costs?
October 20, 2025 at 11:29 AM
A company runs its backend on EC2.
But even with zero logins for weeks, the servers keep running and the bill never stops.
What cloud billing model would let them pay only when their code actually runs, instead of 24×7 idle server costs?
But even with zero logins for weeks, the servers keep running and the bill never stops.
What cloud billing model would let them pay only when their code actually runs, instead of 24×7 idle server costs?
The fastest way to fix a bug? Step away from the code.
The slowest way? Keep staring at it like it owes you money.
The slowest way? Keep staring at it like it owes you money.
October 20, 2025 at 3:30 AM
The fastest way to fix a bug? Step away from the code.
The slowest way? Keep staring at it like it owes you money.
The slowest way? Keep staring at it like it owes you money.
MVP ≠ prototype.
MVP = proof people will actually open their wallets.
MVP = proof people will actually open their wallets.
October 19, 2025 at 12:43 PM
MVP ≠ prototype.
MVP = proof people will actually open their wallets.
MVP = proof people will actually open their wallets.
A small e-commerce site suddenly goes viral and traffic spikes from 10 users to 10,000 overnight.
Their single backend server begins to choke under load.
What kind of architecture could scale instantly without the team touching infrastructure?
Their single backend server begins to choke under load.
What kind of architecture could scale instantly without the team touching infrastructure?
October 19, 2025 at 6:27 AM
A small e-commerce site suddenly goes viral and traffic spikes from 10 users to 10,000 overnight.
Their single backend server begins to choke under load.
What kind of architecture could scale instantly without the team touching infrastructure?
Their single backend server begins to choke under load.
What kind of architecture could scale instantly without the team touching infrastructure?
Software engineering = 2 weeks of glorious greenfield code + 6 months of hacking on top of it while quietly dying inside.
October 18, 2025 at 2:32 PM
Software engineering = 2 weeks of glorious greenfield code + 6 months of hacking on top of it while quietly dying inside.
A student tries to reach a blocked site on the school network.
Direct access is denied, but there’s always a workaround.
What network-layer tool could step in as a middleman, quietly fetching the content on their behalf?
Direct access is denied, but there’s always a workaround.
What network-layer tool could step in as a middleman, quietly fetching the content on their behalf?
October 18, 2025 at 12:45 PM
A student tries to reach a blocked site on the school network.
Direct access is denied, but there’s always a workaround.
What network-layer tool could step in as a middleman, quietly fetching the content on their behalf?
Direct access is denied, but there’s always a workaround.
What network-layer tool could step in as a middleman, quietly fetching the content on their behalf?
Every project starts clean.
Every project ends as a pile of “please don’t make me rewrite this.”
Every project ends as a pile of “please don’t make me rewrite this.”
October 17, 2025 at 12:45 PM
Every project starts clean.
Every project ends as a pile of “please don’t make me rewrite this.”
Every project ends as a pile of “please don’t make me rewrite this.”
An organization wants full visibility
which sites employees visit, how long they stay, and the ability to block unsafe ones without installing software on every device.
What network component could sit in the path and enforce this seamlessly?
which sites employees visit, how long they stay, and the ability to block unsafe ones without installing software on every device.
What network component could sit in the path and enforce this seamlessly?
October 17, 2025 at 3:27 AM
An organization wants full visibility
which sites employees visit, how long they stay, and the ability to block unsafe ones without installing software on every device.
What network component could sit in the path and enforce this seamlessly?
which sites employees visit, how long they stay, and the ability to block unsafe ones without installing software on every device.
What network component could sit in the path and enforce this seamlessly?
A speedy CI/CD pipeline is like fast food with quality control.
Keep your test menu small with mighty, snappy unit tests and quick integrations so your feedback loop is drive-thru fast, not a five-hour sit-down meal.
Read the full article...
Keep your test menu small with mighty, snappy unit tests and quick integrations so your feedback loop is drive-thru fast, not a five-hour sit-down meal.
Read the full article...
October 16, 2025 at 12:44 PM
A speedy CI/CD pipeline is like fast food with quality control.
Keep your test menu small with mighty, snappy unit tests and quick integrations so your feedback loop is drive-thru fast, not a five-hour sit-down meal.
Read the full article...
Keep your test menu small with mighty, snappy unit tests and quick integrations so your feedback loop is drive-thru fast, not a five-hour sit-down meal.
Read the full article...
Engineers need to retire 1,400+ old datasets and move 2,000+ users to a new system. Manual execution would be slow, costly, and error-prone.
What approach could automate this migration while keeping it reliable?
What approach could automate this migration while keeping it reliable?
October 16, 2025 at 3:28 AM
Engineers need to retire 1,400+ old datasets and move 2,000+ users to a new system. Manual execution would be slow, costly, and error-prone.
What approach could automate this migration while keeping it reliable?
What approach could automate this migration while keeping it reliable?
Want to check if your emails pass authentication?
Gmail users: Open any email, click 3 dots, select 'Show Original', look for SPF, DKIM, DMARC results.
Gmail users: Open any email, click 3 dots, select 'Show Original', look for SPF, DKIM, DMARC results.
October 15, 2025 at 12:46 PM
Want to check if your emails pass authentication?
Gmail users: Open any email, click 3 dots, select 'Show Original', look for SPF, DKIM, DMARC results.
Gmail users: Open any email, click 3 dots, select 'Show Original', look for SPF, DKIM, DMARC results.
Imagine millions of people pressing “Play” at the same moment across 190 countries.
Every single click has to be tracked in real time, counted, and processed instantly.
How could any system handle that massive surge of events without breaking down?
Every single click has to be tracked in real time, counted, and processed instantly.
How could any system handle that massive surge of events without breaking down?
October 15, 2025 at 3:32 AM
Imagine millions of people pressing “Play” at the same moment across 190 countries.
Every single click has to be tracked in real time, counted, and processed instantly.
How could any system handle that massive surge of events without breaking down?
Every single click has to be tracked in real time, counted, and processed instantly.
How could any system handle that massive surge of events without breaking down?
In a peer-to-peer system, nodes grab files from nearby peers.
Over time, some files spread everywhere while others nearly vanish.
How could the system balance availability and replication so every file is accessible without wasting storage?
Over time, some files spread everywhere while others nearly vanish.
How could the system balance availability and replication so every file is accessible without wasting storage?
October 13, 2025 at 7:32 PM
In a peer-to-peer system, nodes grab files from nearby peers.
Over time, some files spread everywhere while others nearly vanish.
How could the system balance availability and replication so every file is accessible without wasting storage?
Over time, some files spread everywhere while others nearly vanish.
How could the system balance availability and replication so every file is accessible without wasting storage?