UVIN VINDULA

AI IntegrationNext.jsSupabaseE-commerce

uvin.lk

Next.jsSupabaseEducationCeyPay

CyberArsenal

Next.jsTypeScriptSecurityTools

FreshMart

Next.jsSupabaseStripeSanity

Next.jsThree.jsFramer Motion

View all work→

WEB3 ENGINEER — AI DEVELOPER — BITCOIN EDUCATOR — FULL-STACK BUILDER — BLOCKCHAIN AUDITOR — TERRA LABZ DIRECTOR — 10,000+ TAUGHT — TOKEN 2049 ASIA — PUBLISHED AUTHOR — WEB3 ENGINEER — AI DEVELOPER — BITCOIN EDUCATOR — FULL-STACK BUILDER — BLOCKCHAIN AUDITOR — TERRA LABZ DIRECTOR — 10,000+ TAUGHT — TOKEN 2049 ASIA — PUBLISHED AUTHOR — WEB3 ENGINEER — AI DEVELOPER — BITCOIN EDUCATOR — FULL-STACK BUILDER — BLOCKCHAIN AUDITOR — TERRA LABZ DIRECTOR — 10,000+ TAUGHT — TOKEN 2049 ASIA — PUBLISHED AUTHOR —

Staking.sol

Solidity

1// SPDX-License-Identifier: MIT

2pragma solidity ^0.8.24;

4contract StakingVault {

5 mapping(address => uint256) public staked;

6 uint256 public totalStaked;

7 uint256 public rewardRate = 100;

9 function stake() external payable {

10 require(msg.value > 0, "Cannot stake 0");

11 staked[msg.sender] += msg.value;

12 totalStaked += msg.value;

13 emit Staked(msg.sender, msg.value);

14 }

15}

The code
I write

DeFi staking contract with reward distribution

Tech
stack

Not a list of buzzwords. Every tool here is used in production. Built and shipped for real clients.

Web3 & Blockchain

SolidityDeFi, NFTs, DAOs

Foundry / HardhatTesting, deployment

DeFi ProtocolsAMM, lending, staking

wagmi / Ethers.jsFrontend integration

Smart Contract AuditSecurity review

Layer 2Arbitrum, Base, zkSync

Full-Stack

Next.jsApp Router, SSR, ISR

TypeScriptStrict mode, always

ReactServer + Client components

SupabaseAuth, DB, Realtime, Storage

Tailwind CSSv4, design systems

StripePayments, subscriptions

AI & Tools

Anthropic SDKClaude API, agents

RAG SystemsVector search, embeddings

PythonData, automation, AI

LangChainMulti-agent pipelines

VercelDeploy, Edge, Analytics

PostgreSQLPrisma, raw SQL

Play with
me

Think you can beat me? Pick a game. I dare you.

Latest
Articles

Technical writing on Web3, AI, and building things. 150 articles and counting.

View all→

Layer 2 & Scaling16 min

Fusaka, Blobs and Rollup Economics After the Upgrade

Fusaka activated at a blob target of 6 and a max of 9; its second blob-parameter-only fork, BPO2, took the target to 14 and the max to 21 on 7 January 2026. I sampled 120 consecutive mainnet blocks on 14 September 2026 and the mean was 7.28 blobs per block, with 9 blocks carrying none at all. The blob base fee measured 0.011 to 0.028 gwei, pinned near the reserve price EIP-7918 ties to the execution base fee. At those numbers one blob costs between $0.004 and $0.009, and a three-blob rollup batch costs about one and a half cents. Data availability is not the binding constraint on Ethereum scaling in 2026. Meanwhile the gas limit the Ethereum Foundation said would move toward and beyond 100M is still exactly 60,000,000, and an L1 transfer costs cents rather than the $1.10 a popular fee tracker still quotes. This article gives you the measured numbers, the two RPC calls to reproduce the mainnet numbers yourself, the blob-gas arithmetic in full, an eight-chain L2 execution cost table, and the honest answer to the question those numbers raise: if data is nearly free and L1 is cheap, the case for launching a rollup in 2026 is not economics. It is distribution and control, and it is worth saying so out loud.

Web3 Development17 min

EIP-7702 vs ERC-4337: What the Chain Actually Shows

Account abstraction is about 2% of Ethereum blockspace by transaction count. I sampled mainnet and Base on 14 September 2026 with plain JSON-RPC calls: EIP-7702 type-4 transactions ran 0.8% of mainnet traffic, ERC-4337 EntryPoint calls 1.4%. On Base, EntryPoint calls outnumbered type-4 setcode transactions by roughly 12 to 1 — a comparison that flatters 4337, because setcode fires once per account and an EntryPoint call fires on every operation — and the old EntryPoint v0.6 still carries more Base traffic than the 7702-aware v0.8. The cumulative counters everyone quotes — 5.3 million authorizations, 1.28 billion UserOperations, 66.5 million accounts — are structurally inflated and are not user counts. This article reports the real traffic split, shows the exact RPC calls so you can reproduce the measurement yourself, and then gives a decision framework: when 7702 delegation is the right call, when a 4337 account with a paymaster is, and what each one actually costs you in gas and in infrastructure. Includes the paymaster number nobody quotes — $14.1 million of sponsored gas across roughly 1.01 billion sponsored operations, about 1.4 cents each.

Web3 Development15 min

Building an ERC-7683 Solver: Four Calls, Three Risks

ERC-7683 gives a cross-chain solver exactly four calls to implement: resolve, openFor, fill and settlement. Every intents explainer stops at the sentence 'the user signs what they want and solvers compete', which tells you nothing about what you actually have to build. This walks the real interface surface from the spec at erc7683.org — both order structs field by field, IOriginSettler, IDestinationSettler, and the ResolvedCrossChainOrder return that is the only place your liabilities are made explicit. Then it names the three risks the standard deliberately hands to the implementer and does not price: destination-chain reorg exposure in the window between your fill and your repayment, the finality gap between a two-second fill on Across and a six-day-eight-hour challenge window on Arbitrum, and cross-chain inventory rebalancing, which is the actual cost of running a solver and the reason solver sets stay small. It ends with the volume reality nobody pairs with the interface: CoW Protocol's monthly volume fell from $4.276B in January 2026 to $2.400B in July 2026. This is a maturing interface on a shrinking market.

Web3 Development15 min

Passkey Wallets That Work on Mainnet: Building on EIP-7951

Fusaka shipped a secp256r1 precompile on 3 December 2025 and the coverage was almost entirely PeerDAS and blob counts. EIP-7951 is the line item that matters to wallet developers: P-256 is the curve behind Apple Secure Enclave, Android Keystore and WebAuthn, and before Fusaka, verifying a passkey signature on L1 meant a roughly 300,000-gas Solidity verifier or a trusted relayer, while L2s already had the RIP-7212 precompile at about 3,450 gas. At the L1 base fee I measured on 14 September 2026 — 0.183 to 0.436 gwei — that verifier cost $0.14 to $0.33 per signature, 14.3 times the gas of the plain ETH transfer it was authorising. This is the practical build guide: the gas comparison across nine chains, the precompile call, what WebAuthn actually hands you and why it is not a raw signature, how to pair the precompile with EIP-7702 delegation or an ERC-4337 account, the EntryPoint version split I measured on mainnet and Base, and why recovery is the part that will break your product, not signing.

Web3 Development14 min

MEV and Restaking in 2026: What Actually Happened

No, Ethereum does not have a production encrypted mempool. The flagship attempt — Shutter and Primev's threshold-encrypted mempool for the PBS supply chain, announced December 2025 with mainnet to follow — is in maintenance mode by Primev's own documentation, its settlement explorer decommissioned. Meanwhile the story that proposer-builder separation and TEEs decentralised block building does not survive measurement: Titan built 55.78% of blocks over seven days and 62.87% over twenty-four hours on 14 September 2026, BuilderNet held 10.15%, and the top three relays took 88.3% of all relay payload deliveries between them. The second half of this piece covers the restaking unwind with the numbers the retrospectives get wrong, including the real EigenLayer slashing date — 17 April 2025, not April 2026 — and the real TVL curve from a $22.06B peak to $6.59B. Then the engineering question nobody asks: once slashing is real and enforceable, what is an AVS actually worth paying for? Every figure here is dated and sourced, and the reproduction steps are included.

Web3 Development16 min

Quantus Network: What Actually Shipped

Quantus Network's mainnet started at block 0 on 9 September 2026 from GitHub tag v1.0.0-gm, with genesis hash 0xfb5487c0 and runtime spec_version 152. Nearly every article about it is a rewrite of the same press release. This one is written from the runtime source. The Cargo.toml compiles two ML-DSA parameter sets, not the one the docs claim, and the HD wallet crate is built with ml-dsa-65 only, which makes wallet-derived keys NIST Category 3 rather than Category 5. A transparent transaction is roughly 7 KB because ML-DSA has no public-key recovery, so the 2,592-byte public key rides along with the 4,627-byte signature in every extrinsic. The headline throughput figure does not reconcile with the chain's own block constants. Poseidon2 was not chosen for quantum resistance, and the docs say so plainly. 27% of supply was minted at genesis to insiders. The docs audits page lists four engagements with reports pending while the whitepaper lists seven with links, and the docs page is the stale one. Every number here carries its source and its measurement date.

Layer 2 & Scaling17 min

Rollup Stages: What Stage 1 and Stage 2 Actually Mean

On 14 September 2026, L2BEAT tracked 103 scaling projects. Four were Stage 2 and they held under $1M between them. Six were Stage 1 and they held $29.32B. Seventy-three were still Stage 0 holding $8.48B. That is the sentence missing from every write-up that says the big rollups reached Stage 1 and Stage 2 is next. The highest label on the board is worn only by chains with no money on them, and the label everyone celebrates says nothing about how fast the contracts underneath can change. Base is Stage 1 with single-round ZK fraud proofs and an Exit Window row that reads None — its contracts are instantly upgradable behind two approving parties. Arbitrum One is Stage 1 with an Exit Window row that reads None on its emergency upgrade path and 10 days on its regular one. Same badge, and on the emergency path the same exposure — which is the part every write-up misses. This guide teaches you to read the underlying risk rows instead of the summary label: what each row means, which one decides your users' worst case, how three Stage 1 chains compare row by row, why TVS and TVL are not interchangeable, and what I check before I deploy a contract to any L2. Every figure is dated and linked to its primary source.

AI & Machine Learning17 min

Your Agent Benchmark Number Is Mostly Your Harness

Scale AI's standardised SWE-Bench Pro public set tops out at 61.50% across 731 instances, though even that board is mixed: its top rows ran mini-swe-agent under capped cost, not the board's standard SWE-Agent scaffold at 250 turns with uncapped cost. Anthropic reported 81.2% for Claude Fable 5.1 on the same benchmark name, with no scaffold or subset stated. Both numbers are real. They measure different systems. Terminal-Bench 4.0 is worse — one model, three reporters, three scores: 55.8%, 55.1% and 57.88%. Anthropic changed only the container resources on Terminal-Bench 2.0, held the model and harness fixed, and moved the score 6 percentage points with p below 0.01. Their stated conclusion is that, until resource methodology is standardised, their data suggests leaderboard differences below 3 percentage points deserve skepticism until the eval configuration is documented and matched. This article separates the four things a leaderboard row actually measures — model, scaffold, resource budget and timeout — shows what SWE-Bench Pro, Terminal-Bench 4.0 and tau2-bench each test, explains why Amazon Nova Micro 1.0 outranks a frontier model on a saturated benchmark, and gives you the configuration manifest and the pass^k decision you need to build an eval for your own system.

AI & Machine Learning16 min

Tokens Per Task: The Real AI Cost Model in 2026

Terminal-Bench 4.0's maintainers published the number that kills every dollars-per-million comparison: Claude Sonnet 5 burned 21.6 billion tokens on its leaderboard run against 6.5 billion for Opus 5, and scored 12.4% against 51.8%. The cheaper model per million spent 3.3 times the tokens for a quarter of the completions. Then there is the tokenizer — Claude 4.7 and later produce roughly 30% more tokens for identical text, so Sonnet 5 at $2 against Sonnet 4.6 at $3 is about a 13% real saving, not 33%. Prompt caching was repriced at two of the three major vendors in 2026: OpenAI now charges 1.25x for cache writes with a 30-minute reuse window, Anthropic dropped Fable 5.1 cache reads to 0.025x, and Google alone still bills per-hour cache storage. This article gives the current price tables with measurement dates, the silent cache-miss mechanisms, the exact fields to log for cost per completed task, and the session-runtime SKU that no price table shows.

Sri Lanka Tech Scene17 min

How CeyPay Actually Works: Crypto Payments in Sri Lanka

CeyPay is marketed as accepting 50+ cryptocurrencies. Its own API reference tells a narrower story: the Payments endpoint takes USDT or LKR, auto-converts an LKR amount to USDT at creation, and accepts one of two providers, BYBIT or BINANCE. The response field says currency is always USDT for processing. The create-payment call returns a bybit:// deep link and a checkout page hosted at checkout.bybit.com, so the crypto leg is an internal book transfer inside an exchange rather than an on-chain payment. That is not a trick. It follows from the Central Bank of Sri Lanka's position that resident-to-resident transactions inside the country settle in rupees, which CeyPay's own legal log states in those words. This article traces one payment from exchange balance to CEFTS bank credit, and works the arithmetic in the docs' fee examples, which land at 1.50% and 3.50% against a headline of up to 2%. It covers the FX spread nobody quantifies, the settlement pull with its daily cap, the refund path that exists only on Binance Pay, and the regulation behind the structure.

Web3 Development15 min

Uvin Vindula·September 13, 2026

Quantum-Resistant Blockchains Compared

Most quantum-resistant blockchain comparisons rank projects by marketing claim. This one ranks them on the five properties that decide whether a user's funds are actually exposed: which signature scheme, which NIST security category, whether post-quantum signing covers every transaction from genesis or is opt-in, what it costs in bytes and throughput, and what has shipped against what has been announced. Quantus launched a proof-of-work mainnet on 9 September 2026 where every signature is ML-DSA from block 0. QRL has had hash-based signatures on mainnet since June 2018 but its successor chain has no mainnet date. Algorand shipped native Falcon-1024 accounts in August 2026, and user funds still default to Ed25519. Ethereum's largest exposures are 37 million ETH behind BLS and 15 million ETH behind KZG, neither of which any wallet upgrade can fix, and its own L1 target is 2029. Bitcoin has two draft BIPs and no activation path. Every figure carries its source and the date it was measured.

AI & Machine Learning16 min