Skip to content

G. Karthik Koundinya

Projects

6 case studies with measured results, and 18 more public projects. Every case study ends with what didn’t work.

Case studies

Amazon ML Challenge 2026, Entity resolution
BizMatch
Rank 38 of 32,000+ teams
Finds every real copy of a business across ~10M noisy records and rejects look-alike decoys.
- 38Final rank out of 32,000+ teams in the Amazon ML Challenge 2026
- 0.990Best public-leaderboard macro F0.5, up from 0.921 at the first submission
- 98.75%Candidate recall at ~5 candidates per business (TF-IDF + fine-tuned multilingual-e5)
Read the case study
Personal project, Photographs to 3D
ParallaxOps
Turns 15–60 overlapping photographs into a COLMAP 3D reconstruction on one laptop, then checks the geometry against ground truth.
- 0.31%Camera position error as a share of rig spread, ETH3D terrace vs ground truth
- 0.52°Median camera rotation error on the same scene
- 23/23ETH3D terrace DSLR images registered, 87 s through the platform
Read the case study
Quantum ML research, Neural collapse in VQCs
QNC
A research paper showing that training a variational quantum classifier rotates its states rather than contracting them, and where collapse can still appear.
- ~10⁻¹⁵Relative change when checking the exact-rotation result at 4, 6 and 8 qubits: machine precision
- 2–3Orders of magnitude the measurement-visible projection decays during training
- 52Preregistered predictions across 10 stages, each threshold committed before its metric
Read the case study
HackerRank Orchestrate, Aug 2026, LLM routing
WhatsApp Notification Router
20th of 1,983 submissions
Decides, per user, whether each WhatsApp message (text, image poster or voice note) should notify, digest or mute, and cites why.
- 20Rank out of 1,983 submissions; final score 78.3/100
- 37/40Rows that changed between runs with a fixed seed, before majority voting
- 8/8Impersonator-business and prompt-injection messages muted by deterministic overrides
Read the case study
Razorpay AI Buildathon, COD risk decisions
Nishchay
Decides which cash-on-delivery orders to flag for return-to-origin risk, pricing each decision in the merchant's own rupees.
- ₹72.9LNet vs the merchant's existing rule on 44,288 held-out orders (~₹18.8 lakh/month)
- 0.677PR-AUC on the held-out orders, against a 0.304 no-skill baseline
- 0.986Recall on the same held-out orders
Read the case study
HackerRank Orchestrate, May 2026, Multi-agent RAG
Grounded Support Triage
12th of 1,349 participants with all four submissions
Reads a support ticket, finds the answer in the product docs, checks the citations, then replies or escalates.
- 12thPlace of 1,349 participants with all four submissions, HackerRank Orchestrate May 2026 (12,885 registered, 48 countries)
- 774Documents in the retrieval corpus across HackerRank, Claude and Visa support
- 0–2LLM calls for a typical ticket, thanks to regex fast paths
Read the case study

Everything else I’ve shipped

Public repositories only. Filter by the kind of work.