ICFD-31k
A large-scale benchmark for real-time conversational fraud detection.
IJCAI–ECAI 2026 · AI for Social Good Special Track
Hello 👋
I’m Rishi Ahuja, pursuing a B.Tech. with Research in Information Technology at NIT Jalandhar.
My research focuses on building and evaluating reliable AI systems, with current work in language-model evaluation, temporal retrieval, and conversational fraud detection.
Featured research
ScopeBench-PR
★ Best Paper Presentation
Measuring whether LLM reviewers respect a paper’s stated scope.
GlobalSouthAI · IJCAI–ECAI 2026
Temporal Retrieval
Selective retrieval for time-series forecasting beyond long-context scaling.
ICLR 2026 · TSALM Workshop
Updates and awards
- Presented ICFD-31k at IJCAI–ECAI 2026 in Bremen.
- Received the Best Paper Presentation Award and placed runner-up in the GlobalSouthAI 3MT competition.
- Presented ScopeBench-PR at GlobalSouthAI, IJCAI–ECAI 2026.
- Served as a reviewer for the FSMD Workshop at ICML 2026.
- ScopeBench-PR accepted at GlobalSouthAI, IJCAI–ECAI 2026.
- Awarded an IJCAI–AIJ grant.
- ICFD-31k accepted to the IJCAI–ECAI 2026 main conference.
- Presented retrieval-augmented time-series forecasting research at the ICLR 2026 TSALM Workshop.
- Awarded an ICLR 2026 grant.
- Temporal retrieval paper accepted at the ICLR 2026 TSALM Workshop.
Writing
Towards the Modern Transformer Architecture
Tracing how the 2017 Transformer evolved into the recipes used by modern frontier models — norms, activations, attention variants, RoPE, KV cache, and more.
Jul 2026
You Don't Know WebSockets. Yet.
The WebSocket protocol, handshake to frames, and how bidirectional sockets actually work.
Aug 2025
Go Beneath the Abstraction: Building Interactive UIs with FernKit
Building interactive UIs from scratch in C++ with FernKit’s renderer and widgets.
Aug 2025
About
01 / Background
I am pursuing a B.Tech. with Research in Information Technology at NIT Jalandhar. I study how AI systems can be evaluated reliably against the claims made about them.
02 / Research
My current work spans three projects: ICFD-31k in conversational fraud detection, ScopeBench-PR in scope fairness for LLM-assisted peer review, and selective retrieval for time-series forecasting. Across them, I care about reliable evaluation under realistic conditions.
03 / Beyond Research
I write about machine learning, systems, and building software from first principles. The full archive is in Writing; other work and experience are in my CV.