@cerdo03

Engineering @Temple • Prev-Paytm, Zomato • IIT GOA ‘23

 Delhi
linkedin.com/in/cerdo
Joined May 2020
  • Tweets461
  • Following564
  • Followers288
38Media
Hi guys! I am excited to share that after completing 1 year at @Paytm, I am now joining @zomato as SDE -1. Really excited to work on cool products and to collaborate with such cool folks here at Zomato. Extremely grateful for the opportunity!
23
4
1
376
27,097
From one goal of reducing null searches on Zomato, we built one layer over another and fixed one edge case at a time. So many of Zomato’s greats have had their chance to improve search, and I’m glad I got to add my small bit. zomato.com/blog/search-that-…
1
2
70
1/8 I’ve always struggled with long technical articles and research papers. I want to understand them deeply, but working through them takes me a lot of time.
1
1
103
more replies
5/8 Another inspiration is Paras Chopra, someone I admire a lot. His website, Inverted Passion, feels like following someone’s curiosity across so many topics. It makes me want a space where my own thoughts can wander too. invertedpassion.com/

Inverted Passion

Know what's true and do what's right

invertedpassion.com
2
1
18
The inspiration I mentioned here is @paraschopra. His writing on Inverted Passion helped inspire this little personal corner of the internet.
1
10
7/8 A reminder to keep questioning what I think I know, stay open to changing my mind, and make room for ideas I haven’t considered yet.
1
31
8/8 It’s mainly for me. A way to collect my thoughts, put a little of myself out on the internet, and leave a small digital footprint. Still small, still taking shape: cerdo03.github.io/
1
26
Elon thinks people here will pay 20x just because he's Elon Musk 🤣 People aren’t idiots. Tesla failed miserably here, in case you forgot. A massive 1017 units sold in 12 months. BTW, BYD in the USA 👀
Elon Musk
@elonmuskOct 9
Dear Prime Minister Ambani, Please accept my humble apologies for not realizing that you are the real boss of India. Naturally, you would prefer to maintain your monopolistic exploitation of the great people of India, but would you nonetheless consider allowing Starlink to compete? There are many areas of India, as there are in all countries, with no Internet access, which denies children the opportunity for education and for small businesses the opportunity to sell their goods to a global market. Starlink would change their lives for the better. Thank you 🙏
62
191
7
3,122
62,823
Surya Shukla retweeted
Deepinder Goyal
@deepigoyalOct 9
Replying to @chiraggoyal28
If you add our employees, warehouse workers, and all their families, Eternal directly supports the livelihoods of well over 1% Indians 💁‍♂️.. getting this to 5% by 2030 would be damn cool.
49
223
71
1,712
650,439
1/7 We reduced CPU time per request on one of our highest-volume search paths by 25.3%. At nearly identical traffic, peak ECS capacity dropped by approximately 26%. Here’s how we used Codex Autoresearch to achieve it. 🧵
6
110
7/7 Next: run this in an isolated sandbox using representative, sanitized traffic. Traffic → profile → modify → rebuild → replay → compare → keep or discard. The goal: a repeatable optimization pipeline where engineers review every successful candidate.
9
6/7 Production results from the same three-hour peak window, week over week: • Request volume: −0.36% • CPU time/request: −25.3% • Peak ECS capacity: −26% • p95 latency: −11.2% • p99 latency: −7.8% • Memory/task: stable
30
5/7 The most important part: It also discarded changes. A helper becoming significantly faster did not matter if the complete request stayed flat or regressed. Every change had to preserve responses and improve end-to-end CPU without hurting latency, memory, or allocations.
13
4/7 The retained changes included: • Parsing repeated data once • Removing unnecessary serialization and protobuf copies • Making debug-only parsing lazy • Reducing tracking-data copies • Building indexes in one pass • Reusing lookups, preallocating predictable collections
13
3/7 We ran Autoresearch locally as a Codex skill and connected it to production telemetry through Datadog MCP. For each candidate, it followed a controlled loop: Profile → trace → make one narrow change → test correctness → benchmark end to end → keep or discard.
1
40
2/7 At high scale, tiny inefficiencies become infrastructure costs. Repeated decoding, unnecessary copies, redundant serialization, avoidable allocations, and repeated map operations may look insignificant individually—but execute millions of times across a hot request path.
21
Surya Shukla retweeted
CR Sanchez
@crsanchezxJun 12
La diferencia entre escuchar Spotify… y escuchar Apple Music 🟢 Spotify ・Quiere descubrir lo que te gusta ・Quiere enseñarte canciones nuevas ・Quiere entender tus hábitos ・Quiere acompañarte todos los días ・Te vende descubrimiento ・Te vende personalidad ・Te vende momentos ・Te crea una playlist y parece que conoce tu vida ・Hace que una canción parezca parte de tu historia ・Te dice “Music for every mood” ・Y tú, que solo ibas a escuchar una canción, acabas creando la banda sonora de tu vida 🔵 Apple Music ↓↓
67
60
7
510
1,264,444
Load more