Hi, I’m Aarya Shah.

Designing, building, and keeping AI alive.

Six layers of stack, one orchestrator. I build voice agents, automations, RAG pipelines and the integrations that hold them together. Open to interesting conversations.

50+
AI automations
6
layers covered
3
companies
01

What I do, broadly.

Business-first, technical enough to ship. AI systems only work when they connect to real data, real teams, and real failure modes.
01
AI Systems
Agents, RAG, fine-tuning, turn detection, voice agents, prompt engineering, LLM pipelines.
02
Automation
Make, n8n, workflow debugging, replay scripts, Slack ops, CRM sync, Stripe automation.
03
Product Engineering
Python, TypeScript, Go, Dart/Flutter, backend APIs, app surfaces, admin tooling.
04
Platform
Cloud functions, Kubernetes manifests, internal APIs, Firestore, Snowflake, observability tooling.
02

Toolbox — the honest version.

66 tools · 10 categories
PythonTypeScriptJavaScriptGoDart / FlutterJavaSQLC / C++Shell
ReactAstroTailwindFlutterHTML / CSSViteNext.js
OpenAIAnthropicGroqHugging FaceTensorFlowPyTorchBERTCLIPFine-tuningPrompt eng.
VideoSDKLiveKitWhisperElevenLabsSIPTwilioWebRTCSTT / TTS
LangGraphMCPCustom loopsOpenAI FunctionsTool-use patterns
FAISSQdrantEmbeddingsSentenceTransformersHybrid search
Make.comn8nCloud FunctionsWebhooksGraphQLREST
StripeTwilioSendGridSlackMonday.comNotion
PostgresFirestoreSnowflakeRedis
AWSDockerKubernetesCloudflareGitHub ActionsGit
— Most of this lives behind login walls. The pattern shows up the same in every project.
03

Three roles, same playbook.

Vadodara → Surat → Melbourne
Melbourne, AU · Remote · Full-time
  • 50+ production automations across Make, n8n, Stripe, Monday, Twilio, SendGrid and Firestore — 15+ hrs/week reclaimed in manual ops.
  • LLM agents with human approval flows for customer support and recruitment. Churn was built as a deterministic rules engine instead, because it never needed an agent.
  • Set direction for the internal agent platform (MCP integrations, Langfuse tracing, tool level toggles, least privilege access) and authored the agent migration playbook that answered 14 architecture questions from the founder.
  • Replaced Typeform with an internal forms platform, owning its specification, data model and auth design.
  • Webhook + API architectures for real-time billing, CRM sync and cross-platform data orchestration.
  • Owned automation strategy and tooling decisions across the AI Solutions surface.
Surat, IN · Internship
  • Built and fine-tuned BERT models for semantic end-of-utterance detection in voice agents.
  • Shipped to the AI Voice Agents framework: SIP integration, Twilio transfer, Azure OpenAI, STT/TTS/LLM providers, plugin architecture.
  • Production voice-agent deployments on AWS. Public demo site and developer documentation.
  • n8n + Docker content workflows on AWS — Reddit, X, LinkedIn syndication via REST.
Shipped publicly
Namo Turn Detector
Semantic turn detection so voice agents know when you’re actually done speaking. 23+ languages, sub 30ms on CPU, 3.8k+ downloads.
PR ↗Article ↗HF ↗
VideoSDK AI Voice Agents
Contributions across the public agents framework — voice plugin architecture, SIP, providers, docs and the AI-apps demo site.
Repo ↗Demo ↗Docs ↗
Vadodara, IN · Internship
  • Diamond recommendation chatbot — fine-tuned LLM, regex, prompt engineering.
  • Fine-tuned a generative model for product imagery and integrated OpenAI CLIP for image + natural-language search.
✦ Education & Recognition ✦
B.E. Computer Engineering
Gujarat Technological University
October 2021 — June 2025 · GPA 8.4 / 10
— Selected for —
Amazon ML Summer School 2024