Dipankar Sarkar
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
Organisations
🔬 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.
| Tool | What it does | Lang |
|---|---|---|
| brat | Multi-agent harness for AI coding tools — crash-safe state, parallel execution, one CLI. Orchestrates Claude, Aider, OpenCode, Codex, and others. | Rust |
| fast-litellm | Rust-accelerated drop-in replacement for LiteLLM. 3.2× faster connection pooling via PyO3 with zero config. | Rust/Python |
| fast-langgraph | Rust accelerators for LangGraph — up to 700× faster checkpoint operations and 10–50× faster state management. | Rust/Python |
| gity | Makes large Git repos feel instant. Daemon-based fsmonitor + warm caching turns 8s git status into milliseconds. |
Rust |
| stout | Drop-in Homebrew CLI replacement, 10–100× faster. SQLite FTS5 index, parallel downloads, Ed25519 signed updates, CVE auditing. | Rust |
| fastworker | Background workers for Python apps in seconds, without the complexity. | Python |
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.
| Tool | What it does | Lang |
|---|---|---|
| solscript | Write Solidity, deploy to Solana. Transpiler with full Anchor compatibility — no Rust required. | Rust |
| zig-evm | Experimental EVM implementation in Zig. | Zig |
| stxscript | StxScript-to-Clarity transpiler for Stacks smart contract development. | Python |
| tesseract | Multi-rollup environments on Ethereum — enabling atomic transactions across rollups. | — |
What I'm Building Towards
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.
Background
- 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.