30 days, 11 builds

One real, measured solution per company, built in public over 30 days to land a Summer 2027 internship. Every number on this site comes from the build's own report, with what it does not show stated next to it.

The calendar

Days count down to Day 1 (Day 30 is Nov 1). Each build's results page opens on the day it ships. Status updates from builds.json.

  1. Days 30-28Nov 1-3Live

    Telephony-native voice anti-spoofing for Gnani.ai

    Catching voice clones after the audio has gone through an 8 kHz phone line.

    Codec augmentation holds the line: held-out-codec EER 34.42% with it vs 41.84% without (same LCNN). Best published model, SSL+AASIST: 9.15% clean, 20.97% on held-out codecs.

  2. Days 27-25Nov 4-6Live

    pulse-eot-hi: end-of-turn on Hindi calls for Smallest.ai

    Knowing when a Hindi or Hinglish caller has actually finished speaking.

    Pulse's own end-of-turn is a silence timer plus about 0.33 s: at its default 600 ms it cuts in on 2.0% of real Hindi holds and waits 931 ms; treating every is_final as end-of-turn ends about half of turns early.

  3. Days 24-23Nov 7-8Ships Day 24

    Bolna

  4. Days 22-20Nov 9-11Ships Day 22

    Razorpay

  5. Days 19-17Nov 12-14Ships Day 19

    SigNoz

  6. Days 16-14Nov 15-17Ships Day 16

    Sarvam AI

  7. Days 13-11Nov 18-20Ships Day 13

    AlgoTest

  8. Days 10-9Nov 21-22Ships Day 10

    Eternal

  9. Days 8-7Nov 23-24Ships Day 8

    Swiggy

  10. Days 6-4Nov 25-27Ships Day 6

    Lab0

  11. Days 3-1Nov 28-30Ships Day 3

    Cardboard

About

I am a third-year student at IIT Kharagpur working on ML systems and quant. For this challenge I picked eleven companies, wrote a problem statement for each from public information, and built one solution per company with a fixed evaluation protocol: baselines tuned with the same budget, held-out data scored once, and every withdrawn claim kept on the record.

Each build is labelled REAL or SIMULATED exactly as its repository labels it. Pending runs (GPU jobs, API keys) are marked as pending, never filled in.

How this was built. I wrote the problem statements, designed the experiments and reviewed every result; AI coding agents (Claude Code) wrote most of the code.

On X: @NeedaJobLowk