/uses

The stack I actually build with.

No affiliate links. No sponsored slots. Just the tools that survive on real client work, and the honest reason each one earned its place.

Workflow automation

The core engine that runs everything else. Self hosted so pricing is predictable and workflows are versioned in git.

  • n8n
    Primary automation runtime

    Self hostable on a single DigitalOcean droplet. 400+ integrations, Function nodes for custom logic, LangChain nodes for LLM steps.

  • Make
    Secondary, visual-first flows

    When a client already runs on Make and the migration cost is not justified yet. Great error handling and scenarios UI.

  • Zapier
    First mile connectors

    Occasionally used for niche SaaS connectors that n8n does not have native nodes for.

LLM and AI models

Model choice is per-task. Nothing is picked because it is trendy.

  • Anthropic Claude
    Default reasoning model

    Best long context reasoning and tool use for the price. Sonnet handles the majority of production agent workloads.

  • OpenAI
    Utility model + embeddings

    GPT-4o mini for cheap classification, text-embedding-3-large for RAG, DALL-E for asset generation.

  • OpenRouter
    Model routing and fallback

    One API for every model. Useful for A/B testing and quiet failover between providers.

Voice AI and telephony

Two production paths depending on how much control the build needs.

  • Vapi
    Voice agent platform

    Fastest path from idea to a working outbound or inbound agent. Handles interruption, endpointing, call recording.

  • ElevenLabs
    Text to speech

    Best-in-class voice quality with sub 400ms latency on Flash v2.5. Voice cloning for brand consistent agents.

  • Twilio
    Telephony backbone

    Programmable Voice for the phone number layer under both ElevenLabs and custom stacks. PSTN, SIP, DTMF, recording.

Database and storage

Postgres by default. Supabase gives it a UI, auth, and RLS without running my own hosted panel.

  • Supabase
    Primary database + auth

    Managed Postgres with Row Level Security, realtime, edge functions and storage. The default for anything client facing.

  • DigitalOcean
    Compute + managed Postgres

    Where self hosted n8n and one-off Node services live. Predictable pricing, straightforward snapshots.

  • Cloudflare R2
    Object storage + CDN

    Zero egress fees. Used for call recordings, generated assets and RAG document stores.

Product and SaaS

For the apps I ship as products (push8, ExcelErrorFinder) rather than internal workflows.

  • Next.js
    App framework

    React Server Components, streaming, and the App Router make LLM-powered UIs feel instant. Deployed on Vercel or Cloudflare.

  • TanStack Start
    This site

    Full stack React with file based routing, SSR, and typed server functions. What bishal-paul.com is built on.

  • TypeScript
    Language default

    Every production system I ship is TypeScript. The type system pays for itself the first time an API response changes shape.

  • Tailwind CSS
    Styling

    The fastest way to a consistent, custom looking UI without a design system committee. Paired with shadcn/ui for primitives.

Ops, monitoring and CRM

Where builds get watched after handover, and where clients live.

  • Slack
    Approvals + alerts

    Human-in-the-loop steps route through Slack. Every production system posts a heartbeat and error alerts to a dedicated channel.

  • HubSpot
    CRM most clients run on

    Common CRM target for lead capture, deal automation and pipeline sync in inbound and outbound flows.

  • Notion
    Client docs + runbooks

    Every handover ships with a Notion runbook: architecture diagram, credentials location, on-call steps.

  • GitHub
    Version control

    Every n8n workflow is exported to JSON and committed. No workflow lives only inside a UI.

Developer tooling

The day to day machine.

  • Cursor
    AI-first IDE

    Claude-driven editor. Ships production code faster than VS Code alone for anything larger than a snippet.

  • Warp
    Terminal

    Modern terminal with AI command suggestions and shareable blocks.

  • Raycast
    Launcher

    Custom scripts for the ten repetitive lookups I do every day.

Bishal Paul, AI automation engineer
Bishal Paul
Founder, Erudience
Head of AI, Absolute Intelligence UK

Available for new work

Reply within 24 hours

Building on the same stack?

Get a second set of eyes on your automation.

Thirty minute call. Concrete direction by the end, even if we do not end up working together.

  • Architecture review of your current n8n or Make setup
  • Voice AI feasibility on ElevenLabs plus Twilio, or Vapi
  • LLM cost and latency review across Claude, OpenAI and OpenRouter
Write insteadSee recent builds