Software engineer | Typscript | Golang | AI

Classified
Joined September 2025
Growth isn't found in comfort zones, it's written in error logs.
1
130
The biggest impact AI has had on my life has been the increased frequency between "it's so over" and "we're so back" going through my head
10
6
213
4,583
Last one I promise 😂
186
1,860
160
39,794
629,535
So sick of seeing non-ai startups with .ai domains Is this what .com felt like in '98?
22
2
1
93
4,305
Spent some time debugging a particularly stupid Postgres failure. The API was crash-looping on a migration that added a column with IF NOT EXISTS. The migration runner re-ran every migration on every boot. A later migration dropped that column. Net schema effect: zero.
1
1
15
Except Postgres keeps dropped columns as attisdropped entries. After enough deploys, lists hit the 1,600-column limit. Fixed it with tracked, one-time migrations and a verified table rebuild. The database remembers. Unfortunately.
4
Address tradeoffs out loud : "I chose eventual consistency here because X" beats silence every time.
3
Deep dive on 1-2 components: the interviewer will steer you. Follow their lead.
2
Draft the high-level design: boxes and arrows first. No deep dives yet.
2
Estimate the numbers: QPS, storage, bandwidth. Back-of-envelope math, not perfection.
41
Here's the actual procedure interviewers want to see: Clarify scope: ask about scale, read/write ratio, latency needs. (Skipping this is the #1 mistake; you can't design for a system you haven't defined.)
2
L4 for websockets: L7 CAN hold the connection, but it's paying to track state on every one. L4 just forwards packets. Cheaper at scale.
4
Jitter: same backoff schedule for every client = a synchronized retry storm hitting your recovering service at once.
3
Idempotency keys :"retry with backoff" without one turns a $10 charge into $30.
4
Proximity-Based Services "Design Uber." "Design Gopuff." Anything location-based needs to answer: find me everything within X distance, fast.
3
24
Mistake: reaching for a geospatial index at small scale. Under ~100k items, just scan the list the index overhead isn't worth it until you're indexing hundreds of thousands to millions of entities.
6
Worth noting: most systems aren't querying globally. Proximity usually means local users looking for drivers, restaurants, or items near them, not anywhere on Earth. Design for that.
7
Fix: geospatial indexes PostGIS extensions on Postgres, Redis' geospatial data type, or Elasticsearch with geo-queries. The core idea is dividing the map into regions so you can exclude huge swaths of irrelevant area before you even start searching.
13
Multi-Step Processes Order fulfillment. Onboarding. Payments. Multiple services, long-running, and it has to survive a service crashing mid-flow.
5
Handling Large Blobs Videos, images, docs never route these through your app servers. A 2GB upload has no business passing through your API just to get forwarded to storage.
2
6
The hard part isn't the transfer — it's keeping your DB metadata in sync with what's actually sitting in blob storage. Uploads fail mid-transfer, files get orphaned. Event notifications from the storage service are what keep your app's view of the world honest.
7