# Company assistants that answer from your documents and cite the source. > Enterprise chatbots and RAG assistants that answer questions from your internal documents with citations, respect user permissions and handle Turkish well. Source: https://www.algofact.tech/en/services/ai-ml/chatbot-rag Language: en-US Updated: 2026-10-26 Languages: en-US, tr-TR ALGOFACT builds chatbots and RAG (retrieval-augmented generation) assistants that answer questions from internal knowledge such as procedures, contracts, technical documentation and support tickets, with a citation for every answer. Each user only gets documents they are already allowed to see, the assistant links to the passage behind its answer and it says so when it cannot find a source. It works in English and Turkish and runs in the cloud, on premises or hybrid, with GDPR and KVKK requirements handled at design time. ## Key facts - **What it does:** Sourced answers from company documents - **Languages:** English and Turkish content - **Permissions:** Limited to documents the user can access - **Citations:** Document and section link on every answer - **Channels:** Web, intranet or messaging tools - **Personal data:** GDPR and KVKK handled at design time [1] ## When is an enterprise chatbot the right solution? If employees or customers keep asking the same questions in different words, and the answer is written down somewhere but takes too long to find, a RAG-based assistant is the right place to start. The assistant takes the question, retrieves the relevant document passages, hands only those passages to a large language model and returns the answer with its source. The model is not retrained on your documents, so when a procedure changes, the answer changes with it. If your documents are scattered, contradictory or out of date, the first job is content, not an assistant. No retrieval method turns a wrong procedure into a right one. In the discovery call we look at the state of your content together and, where needed, point out which documents should be cleaned up before the assistant goes live. ## RAG assistant, agent or custom model? These three get mixed up often. What the user expects at the end decides which one you need. ### RAG assistant The user wants information and the answer is written in your documents. The assistant takes no action in your systems; it reads, finds and answers with a citation. ### AI agent The user expects a task to be completed, not an answer: opening a record, updating an order, starting an approval flow. Agents often use a RAG assistant as one of their tools. Link: https://www.algofact.tech/en/services/ai-ml/ai-agents ### Custom model The decision is hidden in your historical data rather than in a document: demand forecasts, risk scores or automatic classification. Link: https://www.algofact.tech/en/services/ai-ml/custom-ai-models ## Where it is used Value shows up fastest where the answer already exists in writing but searching for it eats time. ### Internal policy and procedure assistant Answers questions about HR, purchasing, quality and information security procedures with the relevant clause. A new hire finds the rule they need with a single question. ### Technical documentation and field support Answers from maintenance manuals, product specifications and past incident records. A field technician goes straight to the right section of a long manual. ### Customer support assistant Answers questions from your website or app using published help content, and hands the conversation to an agent with a summary when it cannot find an answer. ### Contract and internal regulation search Finds a specific provision across a contract archive or internal regulations and lists the matching clauses side by side. Interpretation and decisions stay with your legal team. ### Voice and multichannel access The same knowledge base can serve web, messaging and phone channels. For an example of our voice interface work, see CodeFlo on /work: a voice agent platform deployed for clients and clinics in Saudi Arabia. ## How the project runs ### Knowledge source inventory We map which sources are in scope (file shares, SharePoint, wikis, PDF archives, support systems), who can see which document and how documents get updated. ### Real question set We collect the questions users ask most and struggle with most, and mark the correct answer and source for each with your subject matter expert. Every release of the assistant is measured on this set. ### Document processing and indexing Documents are split into passages that keep their heading and clause structure, scanned PDFs go through OCR and tables are processed separately. Semantic and keyword search run together to cope with Turkish suffixes and term variants. ### Answer rules Written rules require the assistant to rely only on retrieved sources, to say when sources are insufficient and to decline topics outside its scope. ### Department pilot The assistant opens to one department first. Users mark answers as helpful or wrong; the cause of each wrong answer is investigated and added to the question set. ### Rollout We open it to all users with single sign-on, logging, usage and cost tracking, and set up automatic re-indexing whenever documents change. ## Access control, citations and data protection - **Document-level permissions:** Retrieval is filtered by the identity of the person asking. Someone who cannot see a document in the source system cannot see it through the assistant either, and permissions are re-read from the source on every indexing run. - **Verifiable answers:** Every answer includes the name, section and link of the document it relies on. The question, the retrieved passages and the generated answer are logged together for audit. - **An assistant that admits gaps:** If the retrieved passages do not cover the question, the assistant does not guess. It says it could not find the information and suggests the team to contact, and this behavior is measured on the question set. - **Document content and rules:** Text written inside a document cannot change the assistant’s rules. Every uploaded file is treated as a knowledge source, never as a command. - **Data residency, GDPR and KVKK:** If documents contain personal data, the requirements of the GDPR or Türkiye’s Law No. 6698 are handled in the indexing design: unneeded personal fields stay out of the index and logs are masked. If data must not leave your environment, the language model and the vector index run on your infrastructure. Legal interpretation stays with your counsel. ## What you receive ### Assistant application A web interface or a chat component embedded in your intranet, messaging tool or customer portal. ### Indexing pipeline A pipeline that pulls documents from source systems, splits them, carries permission data and tracks changes. ### Question set and evaluation report A set of real questions, accuracy and citation results, and a comparison between releases. ### Admin dashboard Questions asked, unanswered topics, user feedback and query cost on one screen. ### Documentation and runbook How to add a new source, how to swap the language model and how to investigate a wrong answer. ## Questions ### What happens if the assistant gives a wrong or made-up answer? The assistant is built to answer only from retrieved sources and to cite them every time. Answers users flag as wrong are reviewed, the cause is found (a missing document, poor splitting, a retrieval miss) and the question is added to the evaluation set. ### Can it handle Turkish and English documents together? Yes. Turkish is an agglutinative language, so keyword search alone falls short. We combine multilingual embedding models for semantic search with stem-based keyword search and measure which combination works best on your documents with the question set. A question in English can be answered from a Turkish source, with the citation pointing to the original. ### Which document formats can you process? Most text-bearing sources: PDF, Word, Excel, PowerPoint, HTML, wiki pages and support tickets. Scanned documents are converted with OCR; accuracy on handwriting and low-quality scans is checked separately during the pilot. ### Will our data be used to train the model provider’s models? If a commercial model API is used, we select enterprise terms under which your data is not used for training and retention is acceptable. If those terms are not enough, we run an open model on your own infrastructure. The decision is made together, based on your privacy requirements and the question set results. ### What drives the timeline and cost? The number of source systems, document volume and quality, the share of scanned documents, how complex the permission model is, the number of channels and where the model runs. We present the scope and the expected running cost in writing after the discovery call. ## Related - [AI, ML and Data](https://www.algofact.tech/en/services/ai-ml) - [AI agents](https://www.algofact.tech/en/services/ai-ml/ai-agents) - [KVKK and ISO 27001 compliant software](https://www.algofact.tech/en/services/technology-assessment/kvkk-iso27001-compliance) - [Discovery call](https://www.algofact.tech/en/contact) ## Sources 1. [Personal Data Protection Law (Law No. 6698)](https://www.kvkk.gov.tr/Icerik/6649/Personal-Data-Protection-Law), Personal Data Protection Authority (KVKK)