# Sal Anvarov > Personal portfolio of Sal Anvarov — full stack developer, founder of the software studio DBS Labs (https://dbs-labs.ai) and Co-founder of Cleanlist.ai, based in Toronto. He presents at a Toronto Meetup on AI topics — how to use different AI techniques to build AI-native software. The site catalogs hands-on case studies (TANNED TO, Pairus, Cleanlist, Hopin Technologies, Flurrish, Hellotax, Natura Market) and long-form writing on software architecture, developer experience, frontend safety, and ML internals. Sal builds production software across the stack: TypeScript / React / Next.js on the frontend, Node.js / NestJS / FastAPI on the backend, with deployments on AWS and GCP. Recent focus: AI-powered B2B tooling (Cleanlist.ai), Chrome extension engineering, and lead-enrichment data pipelines. ## About - [DBS Labs — studio](https://dbs-labs.ai): Sal's software studio. Consulting and product engineering offerings live at [dbs-labs.ai/offerings](https://dbs-labs.ai/offerings). - [Bio](https://www.sal-anvarov.com/bio): Background, work history, and how Sal moved from a CS undergrad in Ontario into AI-powered B2B tooling. - [Credentials](https://www.sal-anvarov.com/credentials): Resume — current role at Cleanlist.ai (Co-founder), prior roles at Oproma, Permion.ai, and earlier positions, plus capabilities and expertise breakdown. - [Contact](https://www.sal-anvarov.com/contact): Reach Sal via Formspree, email, or schedule a chat on Calendly. ## Portfolio case studies - [TANNED TO](https://www.sal-anvarov.com/portfolio/tanned-to): Automated spray tan and skincare studio in Toronto — booking, Stripe memberships, and smart-lock check-in across a web app and native iOS app. Stack: React, React Native + Expo, FastAPI, PostgreSQL, Stripe, Clerk, Seam. - [Pairus](https://www.sal-anvarov.com/portfolio/pairus): Anti-dating app — premium matchmaking that pairs AI matching with expert human matchmakers. Stack: Next.js PWA, React, TypeScript, Supabase. - [Cleanlist.ai](https://www.sal-anvarov.com/portfolio/cleanlist): SaaS platform for cleaning leads and enriching LinkedIn contacts. Stack: Next.js, React, TypeScript, FastAPI, MongoDB. Includes a Chrome extension for LinkedIn extraction. - [Hopin Technologies](https://www.sal-anvarov.com/portfolio/hopin-technologies): Tailored shuttle transportation for daily commutes. Stack: React Native, Expo, Firebase, TypeScript, Samsara. - [Flurrish](https://www.sal-anvarov.com/portfolio/flurrish): Campus-focused social media app for personalized, interest-based experiences. - [Hellotax](https://www.sal-anvarov.com/portfolio/hellotax): VAT compliance automation for multi-channel traders across Europe. - [Natura Market](https://www.sal-anvarov.com/portfolio/natura-market): Canadian e-commerce business specializing in wholesale distribution of perishable goods. ## Open source - [Algo Vision](https://algo-vision.sal-anvarov.com/): Twenty-two interactive visualizations of how algorithms compute, from sorting to Raft consensus. [Source](https://github.com/msanvarov/algo-vision) - [Reference Packer](https://reference-packer.netlify.app/): Packs several reference images into one sheet per subject for AI video on Higgsfield, lints it, and builds the prompt for Seedance, Kling, Wan, Veo or Gemini Omni. [Source](https://github.com/msanvarov/reference-packer) - [Ticket Risk TO](https://toronto-parking-risk.netlify.app/): Scores 0–100 how likely you are to get a parking ticket on your Toronto block at a given hour, from 3.4 million City of Toronto tickets. [Source](https://github.com/msanvarov/toronto-parking-risk) - [gated-attention-minimal](https://github.com/msanvarov/gated-attention-minimal): Minimal PyTorch implementation of gated attention, the G1 variant (arXiv:2505.06708). [Live demo](https://gated-attention-minimal.netlify.app/) - [titans-minimal](https://github.com/msanvarov/titans-minimal): Minimal PyTorch implementation of Titans' test-time neural memory (arXiv:2501.00663). [Live demo](https://titans-minimal.netlify.app/) - [gspo-minimal](https://github.com/msanvarov/gspo-minimal): Minimal PyTorch implementation of Group Sequence Policy Optimization (arXiv:2507.18071). [Live demo](https://gspo-minimal.netlify.app/) - [muon-minimal](https://github.com/msanvarov/muon-minimal): Minimal PyTorch implementation of the Muon optimizer and Moonshot's LLM-scale fixes (arXiv:2502.16982). [Live demo](https://muon-minimal.netlify.app/) - [nsa-minimal](https://github.com/msanvarov/nsa-minimal): Minimal PyTorch implementation of Native Sparse Attention (arXiv:2502.11089). [Live demo](https://nsa-minimal.netlify.app/) - [nodejs-injection-defense](https://github.com/msanvarov/nodejs-injection-defense): Command, SQL, path traversal, prototype pollution and ReDoS bugs in Node.js, each with the exploit, the fix and a test. [Live demo](https://nodejs-injection-defense.netlify.app/) - [nodejs-http-hardening](https://github.com/msanvarov/nodejs-http-hardening): A zero-dependency hardened Node.js HTTP server: headers, body limits, timeouts, rate limiting, CORS, safe errors. [Live demo](https://nodejs-http-hardening.netlify.app/) - [nodejs-supply-chain-security](https://github.com/msanvarov/nodejs-supply-chain-security): Safe npm config, CI auditing, an install-script scanner and the Node permission model. [Live demo](https://nodejs-supply-chain-security.netlify.app/) - [nodejs-crypto-secrets](https://github.com/msanvarov/nodejs-crypto-secrets): Password hashing, secure tokens, JWT verification and AES-GCM encryption with node:crypto. [Live demo](https://nodejs-crypto-secrets.netlify.app/) ## Blog posts - [Pack Your References: Getting Higher-Quality Seedance Videos on Higgsfield](https://www.sal-anvarov.com/blog/seedance-reference-sheets-higgsfield): Packing multiple reference images into one sheet per subject, prompt structure, a two-pass still-then-animate workflow, and shot chaining for consistent Seedance video. - [Practical npm Supply-Chain Defenses for Node.js](https://www.sal-anvarov.com/blog/npm-supply-chain-defenses): A walkthrough of my nodejs-supply-chain-security repo: a safe .npmrc, a least-privilege CI audit workflow, an install-script scanner, and the Node.js permission model, with the tests that prove each one works. - [GSPO Explained: Sequence-Level Importance Ratios for LLM Reinforcement Learning](https://www.sal-anvarov.com/blog/gspo-explained): A practical walkthrough of Group Sequence Policy Optimization (GSPO) from the Qwen team, and of my minimal PyTorch implementation that puts it side by side with GRPO. - [Node.js Crypto and Secrets, Done Right with node:crypto](https://www.sal-anvarov.com/blog/nodejs-crypto-and-secrets-done-right): A walkthrough of password hashing, secure tokens, JWT verification, AES-GCM encryption and secret handling in Node.js using only the built-in crypto module, with tests that show each mistake and each fix. - [Gated Attention, Explained: One Sigmoid After SDPA](https://www.sal-anvarov.com/blog/gated-attention-explained): A walkthrough of the Qwen team paper on gated attention: what a single head-specific sigmoid gate after scaled dot-product attention changes, and a minimal PyTorch implementation with tests that pin each claim down. - [Hardening a Node.js HTTP Server with Zero Dependencies](https://www.sal-anvarov.com/blog/hardening-a-nodejs-http-server): A layer-by-layer walkthrough of a hardened node:http server: security headers, body limits, timeouts, rate limiting, CORS and safe error handling, each one proven by a test. - [Muon, Explained: Orthogonalized Momentum That Scales to LLMs](https://www.sal-anvarov.com/blog/muon-optimizer-explained): A practical walkthrough of the Muon optimizer and the Moonshot AI paper that scaled it to LLM training, built around a ~100-line PyTorch implementation and the tests that pin down how it behaves. - [Five Injection Bugs in Node.js: The Exploit, the Fix, the Test](https://www.sal-anvarov.com/blog/nodejs-injection-attacks-and-fixes): A walkthrough of command injection, SQL injection, path traversal, prototype pollution and ReDoS in Node.js, with the vulnerable code, the fix, and the test that proves each fix works. - [Native Sparse Attention, Explained: A Minimal PyTorch Implementation](https://www.sal-anvarov.com/blog/native-sparse-attention-explained): A walkthrough of the Native Sparse Attention paper from DeepSeek-AI: how compression, selection and a sliding window are combined by a learned gate, and how my small PyTorch reference implementation checks each piece against the paper. - [Titans Explained: A Neural Memory That Learns While It Reads](https://www.sal-anvarov.com/blog/titans-test-time-memory-explained): A practical walkthrough of Google''s Titans paper, where a small MLP memory trains itself by gradient descent during the forward pass, and of my minimal PyTorch implementation of it. - [The Art of Software Architecture](https://www.sal-anvarov.com/blog/art-of-software-architecture): A decentralized approach to software architecture aligned with larger organizational goals and the Advice Process. - [Developer Experience: Beyond the Notion of 'Comfort'](https://www.sal-anvarov.com/blog/devex-beyond-notion-of-comfort): Why DevEx based on what "feels good" can be a poor signal and how to use UX principles to evaluate developer tools. - [Safety Threats in Front-End Development and Mitigation Strategies](https://www.sal-anvarov.com/blog/safety-threats-in-front-end-development-and-mitigation-strategies): XSS, CSRF, and other browser-side threats with concrete mitigations. - [Understanding PyTorch Hooks](https://www.sal-anvarov.com/blog/understanding-pytorch-hooks): How forward / backward hooks unlock advanced debugging in PyTorch without polluting the forward() function. - [Portfolio template for developers](https://www.sal-anvarov.com/blog/free-dev-website-template): The open-source template that powers this site. ## Discovery - [llms-full.txt](https://www.sal-anvarov.com/llms-full.txt): Full text of every blog post and case study in one file. - [rss.xml](https://www.sal-anvarov.com/rss.xml): Blog feed. - [robots.txt](https://www.sal-anvarov.com/robots.txt) - [sitemap.xml](https://www.sal-anvarov.com/sitemap.xml) ## Notes for crawlers Every route is prerendered to static HTML at build time, so the full page text, canonical URL, Open Graph tags and JSON-LD (Person, WebSite, BlogPosting, CreativeWork, BreadcrumbList) are present without running JavaScript. The interactive app renders over that markup in the browser.