I build the systems that keep live video playing — and lately, the tooling that keeps AI agents honest.
- 🎬 Lead Software Engineer at Samsung, Content Engineering team (Samsung TV Plus) — leading and mentoring a team of 8
- 📡 Day job is HLS/DASH live streaming at scale: ABR ladders, DRM, CDN architecture, and real-time channel monitoring
- 🤖 Nights and weekends are multi-agent orchestration — LangGraph, MCP tooling, agent harnesses, evaluation
- ⚡ ~5 years in, C++/Python/Go by trade, and a strong bias toward production-grade over prototype
- 🏆 Samsung Best Growth Project Award, two years running
Three ordinary MCP servers can burn most of your context window on tool schemas alone, before an agent reads a single message. Nothing in the MCP stack warns you. This does.
pip install mcp-token-lint
mcp-token-lint check ~/.config/claude/mcp.json --budget 40000Tokenizes every tool schema across every configured server, reports cost per server and per tool, and fails like a CI budget check when you cross your threshold. Rule-based waste detection, an interactive fix mode, runs fully offline, and refuses to report success when it couldn't measure a server.
| Project | What it is | Stack |
|---|---|---|
| mcp-token-lint | Token budget linter for MCP tool schemas — published on PyPI, CI across 3 OSes | Python MCP tiktoken |
| uniquery | Ask Postgres or MongoDB questions in plain English — MCP server + web UI, with PII masking and read-only guardrails | TypeScript MCP Vercel AI SDK |
| vpp-studio | Virtual product placement & SSAI — multi-model detection, 7-factor placement scoring, PyTorch inpainting | Node TensorFlow.js PyTorch ffmpeg |
| trans-ai | Real-time speech translation for live browser audio — streaming STT → translation → TTS | FastAPI WebSockets Chrome Ext |
| trading-bot | Risk-first crypto framework — strategies propose, the risk manager disposes | Python CCXT XGBoost |
| medconnect | B2B medical marketplace — retailer, wholesaler and admin surfaces | Next.js Prisma Turborepo |
| ide-portfolio | Portfolio styled as a VS Code / Cursor IDE — 4 themes, terminal, command palette | HTML CSS JS |
| expense-tracker-saas | Multi-tenant expense SaaS driven by your own Telegram bot | FastAPI PostgreSQL |
Languages
Streaming & Media
AI & Agents
Backend & Infra
- Building MCP tooling — measuring what agent context actually costs, and trimming it
- Going deep on agent evaluation and observability: eval harnesses, retrieval quality, production reliability
- Writing about streaming internals and agent engineering on Medium
Open to conversations about streaming infrastructure, backend systems, and AI agent engineering.
LinkedIn · Portfolio · Medium · 📫 dhingrayash001@gmail.com