AI agents, chatbots and data pipelines

A chatbot on your website

Ask AI, the analyst inside CrocoJobs, answers in the language of the question (French, English or Arabic), shows its answer as it is written, keeps each conversation as a thread, and cites the postings and datasets it used. A chatbot for your website works the same way, on your products, your documents or your support history.

Trained on your own documents (RAG)

Retrieval-augmented generation (RAG) means the assistant looks things up in your data before it answers, instead of answering from memory. In CrocoJobs the agent plans which tools to call, reads data only through fixed, approved queries (the model never writes its own), and works out comparisons in code rather than letting the model guess.

Its career suite does the same with a person’s own files: upload CVs and letters as PDF, Word or Markdown, and it builds a structured profile from them that the person can still edit.

Semantic search for your site or catalogue

Keyword search fails when the customer uses another word, or another language. CrocoJobs combines classic text search with multilingual semantic search, so an English query finds French postings, and its smart search turns a plain sentence (“accountant jobs in Sfax on a permanent contract”) into the right filters.

Automating work with AI

The best automation is work nobody has to remember. Every night CrocoJobs collects new postings, retires closed ones, enriches and estimates, checks its own data quality and reports what it did. Scoop News translates each new article from Arabic into French and posts it to Facebook through Make.com, so the journalist only writes.

Adding AI to an existing product

An AI feature should not hang on one provider. CrocoJobs sends every call through fallback chains across Groq, Gemini, OpenRouter, Cloudflare Workers AI and Hugging Face, so one exhausted model steps aside without an outage. Every prompt is versioned, every call is traced, and answer quality is measured with a set of test questions and an automatic judge that checks each answer against its sources.

Machine learning models

Some questions need a model trained on your data. CrocoJobs estimates a salary for 10,500+ postings that publish none, with a likely range and a confidence score: on salaries it had never seen, 81% fell inside the predicted 80% range. A new model replaces the current one only if it does better on postings neither of them was trained on.

Data pipelines and dashboards

Most companies already have the data; it sits in exports, spreadsheets and the tools they use. A data pipeline collects it on a schedule, cleans it into one model and keeps it up to date, and a dashboard makes it readable for the people who decide.

Every night, CrocoJobs keeps what it collects untouched, cleans it into one model of a job posting, enriches it, and rebuilds 35 analytics tables behind a statistics space anyone can slice by period, region, occupation, industry, seniority and skill. A data-quality check runs before the numbers change, so a bad night never reaches a chart, and every stage of every run is recorded.

Web scraping and data collection

When the data lives on other websites or in documents, it has to be collected carefully. CrocoJobs gathers postings from seven sources (job boards, feeds and public APIs), fetches only what it has not seen, slows down and retries instead of hammering a site, and builds each source’s terms into its collector: attribution links, the official API where one is required, and how long data may be kept.

Calculi turned the national orientation guide, a 203-page PDF whose score tables existed only as images, into clean data: about 13,000 table cells recovered, every value checked against the rules of the field, and anything uncertain flagged for review instead of guessed.

Prices

Clients in Tunisia (TND)

Europe (EUR)

Rest of the world (USD)

Prices for clients in Tunisia, clients in Europe, clients outside Tunisia and Europe. Each range is where most projects of that kind land; you get a written quote after a first call about what you need. Domains, hosting, paid services and store fees are paid to their providers, in your name.

Proof

• CrocoJobs: Seven sources collected every night, 35 analytics tables, semantic search and a RAG agent with 9 tools, on $0 a month of infrastructure.
• Calculi: A 203-page PDF with its tables stored as images, turned into clean, checked data and 870+ pages.
• Scoop News: New articles translated and posted to Facebook automatically.

Questions

How much does an AI chatbot or agent cost?

Building one usually falls in 1,500 – 5,000 TND for clients in Tunisia, €1,000 – €3,500 in Europe and $1,150 – $3,750 elsewhere. Running it costs model usage on top, paid to the AI provider, and that can be kept very low by design: CrocoJobs runs on free tiers with fallbacks, for $0 a month.

Is my data sent to AI companies?

Only where you decide. In CrocoJobs, CVs and profiles go only to the user’s own AI key or to paid models, never to free tiers, and its privacy page says which processors receive what. The same choice is made with you, for your data.

Can a chatbot give wrong answers?

Any language model can. The design keeps it to what your data says: answers are built from the sources it retrieves and cite them, numbers are worked out in code, and a set of test questions is run again after every change.

Can you build dashboards from our own data?

Yes: from your exports, databases or the tools you already use, to charts your team can filter. It starts with a pipeline that collects and cleans the data on a schedule; the dashboard is the part people see. Data work is quoted after a first call about your sources; smaller jobs are billed by the day: 130 – 200 TND a day for clients in Tunisia, €80 – €110 a day in Europe and $80 – $120 a day elsewhere.

Is web scraping legal?

It depends on the website’s terms, the kind of data and where you operate, so each source is checked before anything is collected. CrocoJobs builds each source’s rules into its collector: attribution links, the official API where one is required, and the retention period a licence sets.

Which AI models do you use?

Whichever fits the job and the budget. CrocoJobs uses Groq, Google Gemini, OpenRouter, Cloudflare Workers AI and Hugging Face, with one multilingual model for search. Every provider sits behind the same interface, so changing models is a setting, not a rewrite.

Tell me about your project

Say what it is for, what you have already and by when. I reply with my questions, or with a time for a first call.

• Email: aliayditn02@gmail.com
• Phone: +(216) 44 674 320