Builder · Systems Architect · OSS Researcher
I build high-performance infrastructure tools in Rust and Zig, with a focus on developer tooling, AI agent infrastructure, and blockchain systems. Based in Scotland, I work across three open-source organisations shipping production-grade tools that replace slow incumbents with fast, correct alternatives.
🌐 dipankar.name · 📧 me@dipankar.name · 🐦 @dipankarsarkar · 📚 Publications
🔬 Neul Labs — Agent-Focused AI Infrastructure
Neul Labs is where I concentrate my work on high-performance tooling for AI agents and LLM-powered applications. The thesis is simple: the Python-native AI stack is too slow for production agent workloads. These tools provide Rust-accelerated drop-in replacements and novel orchestration primitives for the agentic AI ecosystem.
Why this matters for agents: As AI agents move from demo to production, they need infrastructure that can handle thousands of concurrent LLM calls, checkpoint complex multi-step workflows without losing state, and orchestrate multiple coding engines in parallel. Neul Labs tools are the plumbing that makes this possible — Rust where it counts, Python where it's convenient.
🧪 Skelf Research — Systems Research & Developer Tools
Skelf Research is the broader research org, building tools across systems programming, quantitative finance, AI/ML, and constraint solving.
| Tool | What it does | Lang |
|---|---|---|
| sigc | The Quant's Compiler — from alpha idea to production trading strategy in minutes. | Rust |
| numaperf | NUMA-first runtime for latency-critical Rust applications. | Rust |
| fastC | C, but safe and agent-friendly. | Rust |
| polymathy | Rust web service transforming traditional search into an answer engine. | Rust |
| zviz | Container isolation for code you can't trust but have to run. | Zig |
| route-switch | Intelligent LLM routing with automatic prompt optimisation. | Go |
| promptel | Declarative prompt engineering — write once, run anywhere. | JavaScript |
| blogus | Extract prompts from your codebase, version them like dependencies, keep everything in sync. | Python |
| savanty | Natural language to constraint solver — describe optimisation problems in English, get mathematically guaranteed solutions. | Python |
| compere | Intelligent pairwise comparisons — better rankings with fewer votes. | Python |
| mullama | Drop-in Ollama replacement. All-in-one local LLM toolkit. | Python |
| llamafu | Run AI models directly on mobile devices. No cloud, no latency, complete privacy. | Dart |
| liath | A programmable database that speaks Lua. Store data, run queries, build AI workflows. | Lua |
⛓️ Cryptuon Research — Blockchain Infrastructure & Protocols
Research-driven blockchain tooling, from smart contract compilers to cross-chain protocols. Our paper "Towards Universal Atomic Composability" was recognised by a16z's crypto startup school.
The common thread across all three orgs is making fast, correct infrastructure for the next generation of software — whether that's AI agents that need sub-millisecond checkpoint recovery, quant strategies that need NUMA-aware execution, or cross-chain transactions that need formal correctness guarantees.
My current focus at Neul Labs is on the agent infrastructure layer: the tools that sit between LLM providers and production applications, making agentic workflows fast, observable, and crash-safe.
- M.S. Computer Science, Arizona State University · B.Tech, IIT Delhi
- Author of Nginx 1 Web Server Implementation Cookbook (Packt)
- Running Neul Labs · St Andrews, Scotland
Open to research collaborations, advisory roles, and conversations about high-performance AI infrastructure.