Available for new roles · Built with Claude Code
I build AI products that reach real users — end‑to‑end.
Architecture and safety guardrails. Voice pipelines and on-device inference. Payments, OAuth, Chrome extensions, dashboards. Two live SaaS products. A privacy-first family agent. A kid-safe tutor used daily by seven kids. Every project shipped, every project used.
Featured work
Six projects, six different problems.
Jarvis V5
A privacy-first family AI agent that runs on my own hardware.
PostyPop
One piece of content → 22 platform-tuned variations in seconds. Live SaaS, real customers.
Prerna Kapoor — Colorado Realtor
Full website plus a self-running AI content and listings pipeline for a working real-estate practice.
KidsAI Buddy
Kid-safe AI tutor used daily by seven kids — fully local inference, fully local TTS.
Jarvis Mac
A fully local voice assistant on Apple Silicon — streaming pipeline, speculative decoding, on-device RAG.
Mission Control
A homelab dashboard that watches 50 projects and writes me a daily LLM-summarized briefing.
Lead Engine
End-to-end lead pipeline: prospect, audit, generate a preview redesign, send a personalized cold email, track every open and click.
Built with Claude
Every line of this site, written with Claude Code.
The portfolio you are reading right now was built end to end with Claude Code as my pair programmer. So was every other project on this page. The model writes the code. I make the calls.
Several of these projects also ship Claude as part of their production runtime, not just their build process:
- Jarvis V5 routes between local Ollama and Claude Sonnet 4.6 based on per-task privacy classification.
- Lead Engine uses Claude to read each prospect's website, score it against an SEO and UX rubric, and draft the personalized cold email.
- Prerna Kapoor runs three Claude Sonnet workflows live: a site chatbot, a backend listing-creation bot, and a scheduled blog pipeline.
- Foundation Implants was a paid build where every page of copy, schema, and FAQ was drafted with Claude in the loop.
The rest of the work uses local models on principle (KidsAI runs Ollama Gemma so kids' utterances never leave the home network; Jarvis Mac runs mlx-lm Gemma 3.1 entirely on Apple Silicon for the same reason). Picking the right model for the constraint is the job. Claude is the default when nothing else makes sense.
The toolkit
What I reach for
AI / ML
- Anthropic Claude (Sonnet 4.6, Opus 4.7)
- Ollama (qwen3.5, gemma3)
- MLX + speculative decoding
- Groq + multi-model fallback
- pgvector / RAG
- Voice (STT/TTS/VAD, Kokoro, parakeet, Silero)
- Evals & A/B (OpenRouter)
- Prompt caching, system-prompt design
Languages & Runtimes
- TypeScript
- Python
- Node.js 22
- Bun
- Swift / SwiftUI
- SQL
Frameworks
- Next.js 15 / 16
- Astro 5
- React 19
- React Native (Expo)
- FastAPI
- Vercel AI SDK
- Tailwind v4
- Drizzle ORM
Data & Infra
- PostgreSQL + pgvector
- MySQL
- SQLite + sqlite-vec
- Supabase
- Proxmox / LXC
- Docker / Compose
- Cloudflare Tunnel
- systemd / launchd / PM2
Product
- Stripe / Lemon Squeezy
- OAuth (Google, Apple)
- Chrome Extensions (MV3)
- WordPress / Kadence
- SEO + JSON-LD + llms.txt
Also built
Smaller things, useful things, retired things.
Foundation Implants & Periodontics
Premium 40-page WordPress redesign for a Colorado periodontics practice — a paid commercial engagement.
Expense Tracker
A mobile-first PWA my mother uses every day in India.
Market Pulse
Daily market intelligence dashboard — seven tabs, one scheduled LLM aggregation pipeline.
Markdown App (Codex)
A premium native markdown reader for macOS, iPad, and iOS.
Crypto Portfolio
Multi-exchange P&L viewer — and the post-mortem on a retired trading-bot experiment.
Games & Experiments
Interactive prototypes — the place I use to learn things that don't fit in product code.
About
I'm VK. I build AI products.
I build AI products. Privacy-first agents, kid-safe educational AI, production SaaS with real customers, on-device voice — every piece shipped, every piece used.
I work end-to-end: model architecture and safety guardrails through to payments, OAuth, observability, and the unglamorous middle layer that decides whether a real customer stays. The work on this site spans local LLMs on Apple Silicon, multi-agent systems with row-level ACLs, and a paid commercial engagement with a real client.