I believe we're artists before we're engineers. Every piece of complicated code has its own kind of beauty — but there's always a simpler version underneath, and that's elegance. Chasing it is what I actually live for.
The chase, so far:
🔬 Where I was headed — Research. Internship at IBM Research building mathematical benchmarks for LLM evaluation (→ EMNLP-25 submission). Published work on efficiency/checkpointing systems (IEEE-iSES 2025). This was the plan.
🤖 Where I fell instead — Somewhere in the middle of that, I found I loved building more than proving. Took that same instinct for elegance and pointed it at shipping: production ingestion pipelines, agentic LLM orchestration, systems that actually run instead of just getting published.
🌐 Where I tested it fast — Web3 became the proving ground: MVPs across Stellar, EVM, DePIN, and payments infra, a decentralized asset manager (Top 5, Mantle Global Hackathon), and a founding engineer role that raised funding building on-chain systems from scratch.
⚙️ Where I am now — Back in AI, but as an engineer, not a researcher. Building the real, unglamorous plumbing that agentic systems actually run on — and slowly learning that good software design is its own kind of elegance, maybe the truest test of it. Less about proving something works once, more about making it keep working.
⚛️ What's next — Quantum algorithms. Same pursuit, new frontier: what's the simplest, most elegant way to compute this?
Background:
- Research internship, IBM Research (NLP benchmarking → EMNLP-25 submission), published — Light-weight Adaptive Dual-Threshold based Checkpointing for IMCs, IEEE-iSES 2025
- Founding engineer, early-stage startup (MVP → funded beta)
- Currently building production AI infra at ByteBell.ai
Most of my production work is under NDA — not much to show in repos here, but happy to talk architecture and design decisions directly.
📄 Resume 💼 LinkedIn · ✉️ divyraj.saini22b@iiitg.ac.in