Buildoto
Technical details

RAG on documents: pgvector (Supabase) or Vectorize (Cloudflare)?

The RAG pipeline is already built and in production, I adapt it to your context: ingestion, chunking, embeddings, similarity search and cited answers. The choice of vector engine depends on your stack, and also drives the running cost.

Under the hood: RAG · Supabase pgvector / Cloudflare Vectorize. Shipped to production.

SM

Sébastien Mignot

TJM 440 € HT/jour · forfait fixe

assistant.buildoto.com
RAG assistant · your documents
What is the notice period in the Acme contract?
The termination notice period is 3 months, to be given by registered letter (article 8.2).
Source: contrat-acme.pdf · page 4

The problem

What slows you down, and what we change

Today

  • You waste enormous time hunting for the right information across dozens of PDFs, contracts and procedures.
  • Answers depend on a single person who "knows where it is".
  • A generic ChatGPT makes things up about your business and never cites its sources.

With Buildoto

  • The assistant finds the answer in seconds, with the exact source.
  • Knowledge becomes accessible to the whole team, 24/7.
  • Confined to your documents: it cites its source, or says it doesn't know.

Supabase (pgvector), ideal if your data already lives in PostgreSQL: you keep everything in a single database (documents, metadata, vectors, multi-tenant RLS), classic SQL queries + vector search in the same place. The good default for most SaaS.

Cloudflare (Vectorize + Workers AI + R2), ideal for a 100% edge pipeline: upload to R2, extraction and embeddings in a Worker, Vectorize index, low latency worldwide, controlled costs. Relevant when the volume is high or global latency matters.

Both together, common: Supabase for application data and auth, Cloudflare up front for edge ingestion and delivery. I decide case by case based on your constraints (existing data, volume, confidentiality, budget).

The criteria that tip the balance

  • Existing data: already in PostgreSQL → pgvector, with no new component to operate.
  • Multi-tenant: per-client isolation required → pgvector + RLS, filtering is structural, not at the application level.
  • Volume: up to a few hundred thousand vectors, pgvector holds up very well with an HNSW index; beyond several million or in multi-region, Vectorize takes the advantage.
  • Global latency: users across several continents → Vectorize at the edge, no round trip to a single region.
  • Running cost: both stay under a few dozen euros per month for SME usage; the main cost item is the generation API, not the vector database.

The packaged offer

The complete pipeline (ingestion, chunking, embeddings, hybrid search, citations) is already built and runs in production on beforbuild.com. I deploy it on your stack, pgvector or Vectorize, from 2 640 € excl. VAT at a fixed price, live in 2 to 4 weeks:

  • Indexing of your documents and tuning of the chunking on your real content.
  • Sourced answers, a "no evidence, no answer" clause, acceptance testing on your business questions.
  • Integration on your channel (app, website, intranet, Slack) and documentation maintainable by your team.

The details of the feature are on the Chatbot on your documents page.

Deliverables

What you get, in production

01
Reads your documents
PDF, Word, Notion exports, indexed and updated automatically.
02
Sourced answers
Every answer points to the source document; nothing is made up.
03
Integration with your channel
Widget on your site, Slack, Teams or intranet, your choice.
04
Customization & go-live
Tailored to your specifications, deployed and documented, handover-ready for your team.

Proof in production

This feature already runs on beforbuild.com

I designed and deployed this document assistant on beforbuild.com (my B2B SaaS): document search, sourced answers, in production. This is the proven base I adapt to your documents.

Visit beforbuild.com
Sourced
every answer cites its document
24/7
always available
1
point of contact, no subcontracting

Method

Audit → Build → Production

  1. 01

    Free audit · 30 min

    We choose the right feature and scope your specifications. No commitment.

  2. 02

    Fixed-price quote

    Already-built base + customization priced to your specifications. Price and scope signed within 48 h.

  3. 03

    Customization & production launch

    I adapt the feature to your needs, deploy it and document it. Optimized to run at minimal cost.

Fixed price

Base feature

2 640 € excl. VAT

+ customization quoted on your spec · fixed price signed before we start

Or 440 € HT/day on time and materials, adapted to your needs.

Book

Your free audit · 30 min

Pick a slot. We audit your case and you leave with a fixed-price quote.

FAQ

Frequently asked questions

SM

Who I am

A single point of contact, from quote to production

I design, develop and run my own features in production, BeForBuild.com, the B2B SaaS I run on my own, is the proof. These are the already-proven building blocks I offer you, adapted to your needs.

You talk directly to the person who delivers: no subcontracting, no agency layer. The one who designs is the one who codes and ships to production.

Founder of BeForBuild.comA single point of contactFrom quote to production launch

Stack & intégrations

Cloudflare
Supabase
React
shadcn/ui
TypeScript
Mistral
Stripe
Documenso

Let's talk about your project

Book my free audit · 30 min