Applied AI Full-Stack Engineer · Agent Infrastructure, Voice AI & Distributed Systems

Building AI agent runtimes, WebRTC voice pipelines, and high-throughput real-time backends. Active open-source contributor to Google ADK-Go, E2B, and SuperServe.

Portfolio · GitHub · LinkedIn · X / Twitter · Email

A high-concurrency collaborative workspace featuring real-time notes and interactive canvas.

• Built collaborative notes and whiteboard canvas with Yjs CRDTs over WebSockets and Redis Pub/Sub for deterministic state convergence.
• Implemented transactional outbox + Redis Streams chat pipeline with atomic Lua INCR+XADD sequencing and duplicate-write rejection.
• Engineered a TOCTOU-safe AWS S3 vault with transactional quota reservations preventing concurrent upload races.
  Live Demo · Backend · Client

An isolated microVM code-execution sandbox and orchestrator runtime for autonomous coding agents.

• Engineered a Go-based Firecracker runtime configuring ext4 rootfs, kernel, vCPU/RAM, virtio-net, and Unix sockets on bare metal.
• Implemented an in-guest daemon (boxd) communicating with the host over a custom AF_VSOCK HTTP transport for commands, code search, and streaming without exposed guest ports.
• Built TAP/iptables networking with MASQUERADE NAT, outbound HTTP CONNECT proxy, process-group termination, and orphan VM sweeping.
  GitHub

An end-to-end conversational voice AI intake system for high-stakes client triage and consultation.

• Cascaded voice pipeline: LiveKit (WebRTC) audio transport → Deepgram Nova (STT) → OpenRouter LLMs → Deepgram Aura (TTS) with sub-second turn latency.
• Deterministic 4-agent state machine (Triage, Intake, Qualification, Scheduling) where code validates phase boundaries and LLMs execute reasoning.
• Real-time observability: Server-Sent Events (SSE) stream off a Redis event bus into a Next.js live telemetry console.
  GitHub

A low-latency crypto perpetual futures exchange engine built from scratch.

• Hand-written in-memory orderbook matching engine with price-time priority.
• Event-driven pipeline using Redis Streams for durable write ordering and WebSockets for real-time depth/trade fanout.
• Automated PnL and liquidation engine fed by live Binance mark price feeds; 1m/1h/1d continuous aggregates in TimescaleDB.
  GitHub

• Google ADK-Go (5 Merged PRs):
  • Handled multi-part artifact loading when combined with multi-tool calls (#1320)
  • Prevented socket & goroutine leaks on context cancellation (#1351)
  • Deferred telemetry shutdown on launcher server failure (#1470)
  • Added memory store nil guards and allowed valid empty-text MCP tool responses (#1349, #1353)
• E2B Runtime (Merged PR):
  • Implemented non-positive timeout validation across sandbox create, resume, and fork APIs (#3419)
• SuperServe (Merged PR):
  • Resolved quota alert serialization failure to guarantee Slack/email notifications (#324)

• Cartigram — Full-Stack Developer (Dec 2025 – Present)
  Engineered multi-source authorization across 8 ERP modules, built a dynamic approval workflow engine, and implemented CRM lead ingestion pipelines.
• Sheryians Coding School — Full-Stack Developer Intern (Jul 2025 – Oct 2025)
  Built candidate-management APIs, real-time admin dashboards in Next.js, and an automated LangChain-based resume assessment pipeline.

• Languages: Go, TypeScript, JavaScript, Python, SQL, Bash
• Applied AI & Agent Infra: Google ADK, LiveKit (WebRTC), Deepgram (STT/TTS), Firecracker MicroVMs, Agent Sandboxing, LangChain, OpenRouter
• Backend & Distributed Systems: GraphQL Yoga, REST, WebSockets, gRPC, Hono, PostgreSQL, TimescaleDB, Redis Streams, BullMQ, Prisma
• Cloud & DevOps: Linux, Docker, KVM, Kubernetes, Terraform, AWS (EC2, S3, SQS), CI/CD
• Frontend: React 19, Next.js, TanStack Router/Query, Tailwind CSS, Yjs CRDTs, Vite, Bun

Portfolio → itsmesarthak.site
GitHub → github.com/chill-czar
LinkedIn → linkedin.com/in/its-sarthak-chaudhari
X / Twitter → x.com/chill_czar
Email → aisarthak2@gmail.com