Book the audit
All templates
Template · 2 to 3 daysTeams of 5+Ops leadsCustomer supportAgencies

Knowledge-Base Chatbot (RAG)

One workflow does the ingest: pulls Notion pages, Drive docs, PDFs, and 90 days of Slack channels on a nightly cron. Content chunked at 800 tokens with 100-token overlap; OpenAI text-embedding-3-large (3072 dims) writes into Qdrant with per-chunk metadata (source URL, last-updated, ACL tags). A second workflow is the chat loop: Slack @ask mention fires, top-8 chunks retrieved via vector search, re-ranked by Cohere Rerank, and Claude Sonnet answers with inline [1] [2] citations. P95 response: 3.2 seconds. New-joiner ramp time typically halves.

Book the audit · 30 minutes · No pitch

Tool-interchange map · every layer is swappable

Knowledge-Base Chatbot (RAG) · tool-interchange map

01 · IngestNNotion pagesGoogle DrivePDFsSlack channels+3 more
02 · Chunk + embedOOpenAI embeddingsVoyage AICohere EmbedGemini embeddings+3 more
03 · Vector storeQQdrantSupabase pgvectorPineconeWeaviate+3 more
04 · AnswerCClaude SonnetGPT-4oGemini 2.5Mistral Large+3 more
05 · InterfaceSSlack botWeb chat widgetTelegramWhatsApp+3 more

Every layer is swappable · the architecture is not

What it does

Two workflows do the work. The first is the nightly ingest: it pulls your Notion pages, Drive docs, PDFs, and the last 90 days of 3 Slack channels. Content is chunked at 800 tokens with 100-token overlap. OpenAI text-embedding-3-large (3072 dimensions) embeds each chunk and writes it into Qdrant with metadata: source URL, last-updated timestamp, and ACL tags so private content stays private.

The second workflow is the chat loop. A Slack @ask mention or a web chat message fires the pipeline. Top-8 chunks come back from vector search. Cohere Rerank reorders them by relevance. Claude Sonnet writes the answer with inline [1] [2] citations that link back to the source doc. P95 response time sits around 3.2 seconds.

New joiners stop asking the same five questions in Slack. SOPs stop being stale because the source of truth is still the original doc, not a separate export. Ramp time for a new hire typically halves.

Real-world use cases

Who actually uses this, and what changes for them.

Ops lead at a 15-person agency

New joiners ask the same five questions their first week. Senior people spend 3 hours a week answering them.

Before

Onboarding is "read the Notion, ask in Slack, wait an hour for someone senior to answer, move on." SOPs go stale because nobody updates them after they write them.

After

New joiner types @ask in Slack: "what is our handover process for a client workshop?" Answer comes back in 3 seconds with citations linking to the actual SOP doc. If the SOP is stale, the citation is the fastest way for a senior person to notice.

Support team fielding 200+ tickets a week

70% of support tickets are answerable from existing docs, but support reps spend time searching instead of resolving.

Before

Tier-1 support reads the same 12 help articles every day, context-switches between the ticket and Notion, copies half an article into a reply, and hopes it fits.

After

The chatbot drafts the reply from the help centre with citations already inline. The rep spends 20 seconds reviewing instead of 6 minutes searching.

What changes for your team

  • Senior time stops getting eaten by repeat questions.
  • Stale SOPs get flagged naturally because the chatbot surfaces them by query frequency.
  • New joiner ramp time typically halves once the chatbot is available.
  • ACL tags on chunks mean private finance docs stay hidden from junior queries.
  • Answers always come with citations, so nobody trusts the bot blindly.

Real runs, observed

Trigger

@ask "what is our standard contract length for retainers?"

What landed

Answer: "Our standard retainer length is 3 months minimum, extending to 6 or 12 months with a 10% discount respectively [1]. Single-project engagements run 2 to 4 weeks [2]."

Time

2.8 seconds, 2 citations

Trigger

Web widget query from a prospect: "do you support Zoho?"

What landed

Answer: "Yes, Zoho Books is supported as an alternative in the Payment-to-Books template [1]. Zoho CRM is supported as an alternative to HubSpot in the Appointment Router [2]."

Time

3.4 seconds

Default tool stack

NNotion pagesOOpenAI embeddingsQQdrantCClaude SonnetSSlack bot

Sample prompts

Time + scope

2 to 3 days

To ship

Fixed

Pricing model

Included

Source JSON

14 days

Tweak window

Live in these engines

Engine

lock and quay

Internal SOP assistant for the bench

→

Related templates

Template · 2 to 3 days

Appointment Router

Calendly fires the invitee.created event. Clay enriches the email (company, headcount, tech stack, LinkedIn). Claude Haiku picks the right rep from your round-robin config using tier rules (enterprise → enterprise AE, under 10 seats → PLG rep). HubSpot gets the contact plus deal in one upsert. The rep Slack DM lands with the prospect company, 3-bullet context, and the two most recent LinkedIn posts. All inside 18 seconds.

View →Template · 2 to 3 days

YouTube to 5 Platforms

Drop a YouTube URL into a Slack /repurpose command or a Notion row. n8n pulls auto-captions via the YouTube Data API (falls back to Whisper). Claude Sonnet runs 5 separate prompts with platform-specific word budgets: 550-word LinkedIn, 7-tweet X thread, 380-word newsletter section, 180-char Instagram caption with 5 researched hashtags, 75-second TikTok script. Typefully and Buffer each receive their queue via API. Review happens in one Notion page. Publish is one click.

View →Template · 2 to 3 days

Studio Photo Generator

A Drive folder or Shopify admin upload triggers the run. Gemini 2.5 Vision describes each shot in structured form (subject, background type, lighting issues, crop problems). Gemini 2.5 Flash Image re-generates the product on a clean studio backdrop, prompt derived from the vision output. Real-ESRGAN upscales to 4096px. Output files named {sku}-{variant}-{hash}.webp write back to Shopify product media via the Admin API, or to Drive. 200 product shots in 40 minutes wall time at about $6 in model spend.

View →

Ready to ship this?

The audit takes 30 minutes.
The first template ships in days.

Book a call. I review your stack, confirm the fit, and we start the same week.

Book the audit · 30 minutes · No pitchSee all 12 templates →