Hi everyone,
I’m building an AI quoting agent for plumbing/HVAC companies in French-speaking Switzerland. Looking for an experienced n8n developer for a paid, long-term collaboration.
What the agent does:
Takes a free-text intervention request in French
Matches products to exact supplier catalog references (Getaz Miauton, Sanitas Troesch, Nussbaum)
Flags missing products instead of inventing prices
Generates a structured PDF quote with materials, 15% margin, labor (95 CHF/h), travel (45 CHF), and Swiss VAT 8.1%
Non-negotiable: exact catalog matching only. No approximation, no hallucinated references or prices.
Stack: n8n workflows, LLM integration, PDF generation.
Selection: based on a working prototype I can test myself. No prototype = no conversation.
Interested? Contact me directly on WhatsApp: +33 6 12 46 26 13.
Hey 
,
I’m Milan, with 8 years of experience in Business Automation and AI. Including 2 years at Apify working on enterprise-level browser automation.
Currently specializing in n8n, but also proficient in Python & Javascript.
Find out more about my work here:
If you think I might be a match, please:
Book a call here with me
Or reach out at hello@smoothwork.ai
Looking forward to hearing from you!
Hi! I’m an experienced n8n automation developer with hands-on experience building AI workflows, LLM integrations, API connections, and automated PDF generation.
I understand the importance of exact catalog matching with no hallucinated references or pricing, and I can build a reliable prototype that follows your requirements. I’m interested in a long-term collaboration and can start immediately.
I’ll send you a message on WhatsApp right away so we can discuss the prototype and get started.
Looking forward to working with you!
"Hi! I’m reaching out regarding your n8n AI quoting agent role for French-speaking Switzerland.
I specialize in n8n workflows and building hallucination-free LLM integrations using strict RAG/vector matching for product catalogs.
I have a working prototype ready for you to test that handles the free-text French parsing, exact catalog matching, and the 8.1% VAT / pricing structure you outlined.
Let me know if I can drop the test link here!"
Interested. I’d like to know more about what you’re building.
I may be able to help.
contacto@solvo.uy
Hey, I have messaged you on whatsapp. Please check and let me know.
The hard part here isn’t the n8n plumbing, it’s your non-negotiable: exact catalog matching with no invented prices. That’s exactly how I build.
For Getaz Miauton, Sanitas Troesch and Nussbaum I’d index each catalog into a keyword + vector store keyed on the exact reference, and the agent only ever returns a line if it maps to a real reference above a confidence threshold. Anything it can’t match with certainty is flagged as “missing / needs human”, never approximated. The LLM reads the free-text French request and proposes candidates, but a deterministic layer validates every reference against the catalog before it reaches the quote, so a price can’t be hallucinated. That fail-closed principle is literally what I built Fidacy on, an AI system that refuses rather than guesses.
The PDF is the easy part: structured quote with materials, 15% margin, labour at 95 CHF/h, 45 CHF travel and 8.1% Swiss VAT.
I get that it’s prototype-first, no prototype no conversation. I’m happy to build a small working proof on a couple of your real references so you can test the matching and the fail-safe yourself. Quick question that decides how clean the exact matching can be: what format are the supplier catalogs in (PDF, CSV, or an API)?
Hi Milan,
We are Bravio: a two-person team. James handles process and product design; Nick builds as a full-stack developer and fractional CTO.
For Memorians.nl, we integrated an LLM directly into their daily workflow, from technical setup to team adoption. That work is close to what you need here.
Nick works with n8n regularly and can build the quoting agent end-to-end: intake parsing, LLM-driven pricing logic, output formatting, and any back-end connections your workflow requires. We deliver a working system for a fixed price.
We would like to understand your current quoting process, the data sources involved, and what a finished result looks like for you. Happy to set up a short call.
More about our work: bravio.nl
Kind regards,
James Doeland
W www.bravio.nl
T +31 6 5 33 33 531
E james@bravio.nl
Hi! I’m Misha, an automation and LLM integration expert. I’ll build an n8n workflow that processes free-text plumbing/HVAC requests in French, validates every material against the exact supplier catalogs (Getaz Miauton, Sanitas Troesch, and Nussbaum), automatically flags any unmatched items instead of generating guesses, calculates pricing with your required 15% margin, labor (95 CHF/h), travel (45 CHF), and Swiss VAT (8.1%), then generates a clean, structured PDF quote ready for delivery. I can provide a fully working prototype you can test yourself before moving forward, ensuring the workflow meets your exact catalog-matching requirement with zero hallucinated references or prices. This approach gives you a reliable AI quoting system that dramatically reduces manual work, eliminates costly pricing mistakes, scales effortlessly as your business grows, and provides the consistency and accuracy needed for production use—exactly the kind of automation built for long-term success.
Connect on WhatsApp: +375293761570 | View My Portfolio: https://mikedevai.netlify.app/
Hi AD,
I can help with this, but the key part is the catalog-matching design, not just the LLM prompt.
For an exact-match prototype I would keep the first slice very small and testable:
- 5-10 real French intervention requests,
- a limited catalog extract from Getaz/Sanitas/Nussbaum,
- deterministic matching rules first, LLM only for request parsing/classification,
- explicit “missing product / needs human review” output when no exact reference is found,
- generated quote fields for material, margin, labor, travel and Swiss VAT.
I would not let the LLM invent supplier references or prices. The workflow should produce either a confirmed catalog reference or a clear exception queue.
If you can share anonymized sample requests and a small catalog sample, I can scope a small paid proof-of-matching prototype before a larger long-term build.
Portfolio/proof: SHAL Engineering Portfolio - S23 Ultra Proof, Backend, Android, AI API Rescue
GitHub: ashalafan (shalafan) · GitHub
Hi,
This is an interesting project, and I like the emphasis on deterministic catalog matching rather than letting the LLM invent references or pricing.
For this type of workflow, I’d keep supplier catalogs as the single source of truth, use the LLM only for interpreting the customer’s request, then validate every matched item before generating the final PDF quote. Any low-confidence or missing matches should be flagged for review instead of guessed.
I’ve built production n8n workflows involving AI agents, multi-step decision logic, APIs, validation layers, and document generation, so this aligns well with my experience.
I’ve sent you a WhatsApp message to introduce myself and share my portfolio and recent project references.
Looking forward to speaking with you.
Hi,
Your project sounds very interesting, and it aligns well with my background.
I have experience designing and implementing production-grade n8n workflows, integrating LLMs, OCR, databases, external APIs, and document generation. I also have experience building automation pipelines where data integrity and validation are critical, so I completely agree that exact catalog matching should take precedence over any AI-generated approximation.
I’d be interested in learning more about your current architecture and the scope of the prototype you’re expecting. If we’re aligned on the requirements, I’d be happy to build a proof of concept for you.
Feel free to send me more details here or on WhatsApp.
Best regards,
Paulo
https://www.linkedin.com/in/paulo-lima-228a1a179/?skipRedirect=true
Hi AD-1, I build production n8n workflows with LLM/PDF-generation logic (255+ workflows running in production on my own instance, and I use Claude Code as part of my dev workflow which helps me move fast on strict rule-based logic like your “no invented prices” requirement). I get why you want a working prototype before talking — send me the catalog format/sample SKUs and the exact quoting rules and I’ll put together a small working demo against real logic (not a mockup) so you can test it yourself before we talk pricing. Happy to follow up on WhatsApp once I have that.
For the exact-match requirement, I would keep catalog matching deterministic rather than letting the model invent references or prices.
A safe flow is:
- parse the French intervention request into normalized material lines;
- match only against supplier reference fields;
- send unmatched lines to an exception queue with no reference or price;
- apply margin, labor, travel, and VAT only after matching.
Which supplier source is available first: CSV, PDF, or API? A sanitized schema plus 5-10 representative requests would be enough to validate the matching logic.
I have now put the exact-match control path into a live synthetic prototype: tinyopsstudio dot com slash swiss-hvac-quote-prototype
The demo parses a French intervention request, matches only against catalog aliases, leaves unsupported lines unpriced in an exception queue, and applies the 15% material margin, labor, travel, and 8.1% VAT only after a match. It also includes the downloadable n8n workflow. All demo references and prices are intentionally fictitious.
For a production pilot, I can connect one supplier source, validate 10 anonymized requests, generate the structured PDF quote, add failure logging and tests, and deliver the workflow plus handoff notes for CHF 1,500 fixed within 5 business days.
Which supplier source is available first: CSV, PDF, or API?
Your non-negotiable is the interesting part of this build, and it is the part most people will get wrong. Matching a catalog reference is easy. Refusing to match one is the hard bit, because the model would always rather guess than return nothing.
I built that exact behaviour this week for a different domain, store support rather than quoting, and put it on a public page so you can test it instead of taking my word for it: https://citedanswers.netlify.app
The 68 second video shows three things in order. It answers a question and cites the source document. Then it is asked something the documents do not cover, and it says so instead of inventing an answer. Then it hands the conversation to a human and the alert actually arrives. Enforcement sits in two places, a similarity threshold on retrieval and a second check on the generated answer, because one layer alone leaks.
For your case that maps to: catalog reference matched above threshold, or flagged as missing and sent to a human. Never a plausible looking reference that does not exist. Prices come from the catalog rows themselves, never from the model.
One thing to be straight about, I do not speak French. The retrieval and matching layer is language agnostic and I can build and prove it against your catalogs, but if you need someone who can read the French intervention text themselves and judge nuance, I am not that person and you should know before we spend time.
If the approach looks right I will build the matching and refusal slice against a sample of your Getaz or Sanitas data so you can test it yourself, which is what you asked for.
Bonjour AD-1,
Your hardest requirement — “flags missing products instead of inventing prices” — is the exact principle my systems are built on, so let me show the architecture instead of promising it:
1. PARSE (LLM, French): free-text request → structured list of line items (product, qty, unit). The LLM’s job ends here — it never touches prices.
2. MATCH (deterministic, no AI): each item → exact lookup against the supplier catalogs (Getaz Miauton / Sanitas Troesch / Nussbaum). Found → catalog price. Not found → the line is FLAGGED for human review. By construction, a price can’t be hallucinated: there is no code path for the LLM to write a number into the quote.
3. CALC + PDF (plain math nodes): 15% margin, 95 CHF/h labor, 45 CHF transport, 8.1% VAT → structured PDF quote.
I run 33 production n8n workflows daily — AI intake agents and self-checking pipelines where the part that generates is never the part that verifies (that’s how I catch errors before the client does). Portfolio: Vladyslav Tomenko — websites that work + business automation: bots, integrations, reports
Re: prototype — agreed, that’s the right filter. I’ll have a working demo (French text in → PDF out, including the flagged-missing-item case on a sample catalog) ready within 48h. One question to make it realistic: are supplier catalogs available as CSV/Excel/API exports, or should the prototype run on a sample structure first?
WhatsApp’ing you as well. — Vladyslav
Hi AD-1,
This is a strong fit. The key constraint you called out - exact supplier catalog matching and no hallucinated references/prices - is the right one. I would not treat this as a generic chatbot; I would build it as a controlled quoting workflow with deterministic catalog lookup, confidence/missing-product flags, and human review where needed.
A good first milestone could be:
- take a small supplier catalog slice and a few real French intervention requests;
- build the n8n flow for product matching with exact references, confidence scores, and explicit missing-product outputs;
- generate a structured PDF quote with material lines, 15% margin, labor, travel, VAT, and a review fallback;
- document how the workflow should be expanded to more suppliers and edge cases.
My background is in AI automation, Python/API integrations, document workflows, and production workflow design. I can start immediately and work async/remotely. Portfolio: https://charliiai.com
Best,
Charlii Liu