Head of Engineering @LimeChainHQ

Joined December 2009
LimeChain’s audit agent scored 75/75 on the ETHEVALS concepts, transactions and security pillars, beating the best frontier model run. It runs entirely on our own infrastructure with open-weight models, so client code never leaves our network.
On the last Solana audit, our edge came mostly from frontier models refusing. This time it's pure quality - LimeAgent, our closed-alpha agent on open-weight models and our own infra: 75/75 on ETHEVALS concepts, transactions and security. Best frontier run: 72 🧵
1
8
261
On the last Solana audit, our edge came mostly from frontier models refusing. This time it's pure quality - LimeAgent, our closed-alpha agent on open-weight models and our own infra: 75/75 on ETHEVALS concepts, transactions and security. Best frontier run: 72 🧵
We're building a security audit agent that runs entirely on our own infrastructure using open-weight models. A recent Solana audit showed why that matters🧵
1
6
1
12
425
Ognyan Chikov retweeted
Say GM to GoFin! GoFin is a CLI that gives your AI agent the ability to trade, bridge and earn crypto for you. You say what you want, your agent does it, you approve with Touch ID
9
11
4
30
1,480
We're building a security audit agent that runs entirely on our own infrastructure using open-weight models. A recent Solana audit showed why that matters🧵
5
4
1
14
565
We’re joining @Mastercard Innovation Day in Slovenia this week 🇸🇮 Alex Vankov (Product Owner @ LimeChain) will take part in a panel discussion on: “Are Stablecoins Becoming the New Payment Rails?”
1
2
11
311
GLM-5.3 Flash on 2x DGX Spark is getting really fast. 368 tok/s aggregate with 8 concurrent sessions (46/session), ~99 tok/s single-session, 1M context per request. Built on @jayleaton's TensorFold kit with our C8 tuning + an OpenAI logprobs patch. Benchmarked against vLLM (EXL3 and NVFP4) on the same harness. Thanks @jayleaton @miaai_lab github.com/LimeChain/glm-5.3…
2
4
195
At @LimeChainHQ , we built a Web3 workstation for @dhh’s Omarchy ⚡ Pinned EVM stack 🧪 Account-free Anvil 🟣 Offline Surfpool via @txtx_sol + @solana_devs 🖥 Native Quattro panel 🔐 No wallets, keys or signing ⛓ More networks are coming github.com/LimeChain/omarchy…
1
6
1
13
703
GLM-5.3-Flash on 2 DGX Sparks 101.74 aggregate tok/s at C8, our most stable production configuration so far github.com/LimeChain/glm-5.3…
3
9
201
Introducing LimeForge, a new product by LimeChain. Most onchain product ideas don't start as specs. They start as a sentence: “I want to build a lending protocol” “A marketplace for tokenized assets” “A new way for institutions to move assets onchain” The hard part is turning that idea into something a team can actually build. For the last 9 years, we've been doing exactly that at LimeChain. Across hundreds of blockchain products, we've accumulated a pretty ridiculous amount of knowledge about product definition, requirements, architecture, user flows and the decisions that need to happen before development starts. A lot of it lives in our internal docs, specifications, architecture decisions, project retrospectives and the heads of our engineers. So we packaged all that knowledge into LimeForge and we’re giving it to you. You bring it an idea, LimeForge helps you work through the product, define its requirements, map the architecture and user flows, and turn it into a build-ready spec and visual prototype. LimeForge is now live in early access. Link below 👇
2
3
13
645
Ognyan Chikov retweeted
Running DeepSeek-V4-Flash-0731 on 8x3090s is a really good experience. 95 tok/s single-session and up to 400 tok/s aggregate with 12 concurrent on a @vllm_project fork using PP8/TP1 and DSpark. Maybe the RTX 3090 is still better value than DGX Spark? github.com/LimeChain/vllm
2
1
1
11
1,716
You cannot call yourself @BulgarianDegen and not attend Solana Summit Serbia 🇷🇸 Meet Hristo Dimitrov on August 26-27 at Sava Centar, Belgrade or at the nearest shisha joint
I'll be at Solana Summit Serbia 🇷🇸 A two-day conference in Belgrade hosted by @SuperteamBLKN, with government, finance, big tech, investors, and the whole @solana ecosystem, plus plenty of side events. Claim your free ticket 👇 solanasummit.rs
4
1
6
459
Solana Summit Serbia is around the corner 🇷🇸 Before we head to Belgrade, we're doing a lil Solana Summit warm-up mixer in Sofia next week together with our friends at @SuperteamBLKN If you're in town, join us Thursday (July 30) for a @solana Balkan summermaxx evening sesh 🇧🇬
3
1
1
4
732
Happy 1000 epochs @solana 🥳 Congrats to the validators, builders, devs and users who have helped bring Solana to this milestone We’re grateful to be a small part of this journey through our LimeChain validator and continued involvement in the Solana ecosystem Accelerate🏎️
a thousand epochs ago, it was an experiment a thousand epochs from now, they'll say it was obvious
1
1
4
593
Our Solana program step-debugger, Gimlet, has a new release 🚨 It now supports the sbpf-debugger feature from the Agave v4.0 runtime, letting @solana_devs debug @solana programs inside Mollusk/LiteSVM/TypeScript tests Here's an example of debugging a CPI 👇
1
2
14
894
All smiles from the LimeChain mentors team at the @RaylsLabs Hackathon 🇫🇷
1
5
21
923
The stage is set 🤩 On the agenda: 14:30 - "Privacy as an Infrastructure" joined by LimeChain PO Alex Vankov 16:00 - "Stablecoins and Tokenised Deposits" moderated by LimeChain’s CEO @NickyTod
1
6
543
Prop AMMs approach onchain execution differently - prioritizing execution quality over efficiency, often at ~36% higher compute units We’ve built tools that benchmark and analyse prop AMMs, here’s what we found: Prop AMMs are blooming on Solana, with newcomers showing up every month or so. This fundamentally different approach to onchain markets is still underexplored. We’ve taken the liberty of building tools, analysing the most prominent approaches and benchmark performance on different axis -- not only price-wise. For example, different Prop AMMs take different approaches -- some favour fast execution & minimal CU consumption, whilst others prefer to deliberately overconsume, or throttle, execution, ensuring that swaps are positioned further back in the block - the minimalistic oracle price updates; We’ve also measured the significance of the Jupiter overhead, or said differently - how much more costly are the swaps, in terms of compute units and execution time, when they’ve been done through Jupiter rather than directly going through the AMMs/props; On aggregate, swaps take roughly 36% more CU, or execution time, when done through Jupiter rather than direct instruction The following diagrams depict two 5-leg swaps, each executed in a local simulated environment with preloaded state; • the AMMs include: raydium_cp / whirlpool / meteora_damm / manifest / meteora_dlmm • the pmms include: bisonfi / humidifi / goonfi / solfi / tessera the AMM path is ~30.4% faster (as labeled), or put another way, the PMM path costs ~30.4% more. Effectively meaning that prop AMMs seem to lean on the “heavier” side -- effectively getting distributed further back in the block and ensuring no adverse selection happens by getting swapped against with stale state
3
1
8
1,981
this is super good btw!
We’ve released sbpf-coverage - a tool designed to generate testing code coverage for @solana programs At LimeChain we believe testing code coverage is an important part of the development lifecycle and something the @solana_devs community has been asking for It's finally here🧵
2
18
3,378
Builders at EthCC, this one’s for YOU @RaylsLabs is hosting an onsite hackathon during @EthCC Cannes 🇫🇷 Build the full lifecycle of institutional assets on-chain: from issuance to privacy-preserving automation. 🗓️March 28-29 📌Château on the French Riviera 💰$17,500 prize pool
3
2
2
11
5,344
You can build something great in Web3 and still get ignored It’s not because the product is bad But because nobody understands why it matters We’re doing free online consultations, where you can talk with @BilyanaChristov about sharp inputs + clear next steps for your project
1
1
4
255