How to Build an Enterprise AI Assistant in Microsoft Teams with RAG

Employees spend up to 30% of their working time searching for information in SharePoint and communicating in Teams. For a company of 1000 people, this amounts to tens of thousands of unproductive hours per year — substantial losses for the business. The ticket system is overloaded with repetitive HR, IT, and document management questions. An AI chatbot embedded in Microsoft Teams solves both problems: it answers queries directly in the chat using the corporate knowledge base. We develop such bots turnkey — from architecture design to tenant-wide deployment.

At the core is a hybrid RAG architecture: an LLM (GPT-4, LLaMA 3, or Mistral) generates responses, and a vector database (Qdrant, ChromaDB, pgvector) stores 1536-dimensional document embeddings. This provides up-to-date answers without retraining the model, with hallucination control through chunk-based search. The 128K token context window handles entire documents, and chain-of-thought prompting improves complex query quality.

What problems does the Teams bot solve?

Ticket overload: 40% of helpdesk queries are routine (password reset, ticket status, policies). The bot responds instantly, offloading first-line support by 60%. The savings on helpdesk are substantial for medium-sized businesses, potentially saving over $100,000 annually for a 1000-employee company.

• Slow document search: An employee spends on average 12 minutes finding the right file in SharePoint. The RAG bot finds an answer in 2-3 seconds using semantic search on embeddings.
• No single entry point: Information is scattered across Outlook, Teams, OneDrive, and internal portals. The bot aggregates data via Graph API and Adaptive Cards, displaying results as interactive cards.

Why RAG over fine-tuning?

Fine-tuning a model on corporate knowledge requires labeled data, computational resources, and risks overfitting. RAG supplements the prompt with relevant documents at query time — no retraining, guaranteed relevance. In practice, RAG architecture shows 60% higher accuracy (F1 metric) on frequently updated documents compared to fine-tuning. RAG on Wikipedia Additionally, RAG allows changing data sources quickly without restarting the bot, whereas fine-tuning requires retraining.

Architecture of a Teams bot

The Teams bot is built on Azure Bot Framework. The platform supports automatic scaling and deep Teams integration. Key components:

• Bot Framework SDK (Python/C#/Node.js) — bot logic
• Azure Bot Service — registration and routing
• Bot Framework Connector — Teams integration

from botbuilder.core import ActivityHandler, TurnContext
from botbuilder.schema import ChannelAccount, Activity

class MyBot(ActivityHandler):
    async def on_message_activity(self, turn_context: TurnContext):
        user_input = turn_context.activity.text
        response = await ai_handler.process(
            user_input,
            user_id=turn_context.activity.from_property.id
        )
        await turn_context.send_activity(Activity(type="message", text=response))

    async def on_members_added_activity(
        self, members_added: list[ChannelAccount], turn_context: TurnContext
    ):
        for member in members_added:
            if member.id != turn_context.activity.recipient.id:
                await turn_context.send_activity("Hello! How can I help?")

Adaptive Cards

Teams Adaptive Cards are JSON descriptions of UI components. They are far richer than text messages:

{
    "type": "AdaptiveCard",
    "$schema": "http://adaptivecards.io/schemas/adaptive-card.json",
    "version": "1.5",
    "body": [
        {"type": "TextBlock", "text": "Analysis Result", "weight": "bolder"},
        {"type": "TextBlock", "text": "{{analysis_text}}", "wrap": true}
    ],
    "actions": [
        {"type": "Action.Submit", "title": "Accept", "data": {"action": "accept"}},
        {"type": "Action.OpenUrl", "title": "Details", "url": "{{details_url}}"}
    ]
}

Integration with Microsoft 365

The Teams bot can access Graph API with the user context:

• SharePoint: search documents, retrieve files
• Outlook Calendar: create meetings, check availability
• Azure AD: org structure, find colleagues, groups
• OneDrive: access user files

Deployment approach comparison

Vector database comparison

Qdrant is up to 2x faster than ChromaDB and 3x faster than pgvector, making it ideal for high-throughput enterprise scenarios.

What's included in the work

1. Analysis and design: audit of existing data sources, choice of RAG scheme (hierarchical or flat), definition of intent classes.
2. Development: implementation of core bot logic, Graph API integration, vector database setup, prompt engineering calibration (few-shot, chain-of-thought). For a corporate assistant in Teams, we tailor the logic to your business needs.
3. Testing: simulation of 100+ parallel requests, latency p99 check (< 2 seconds), A/B testing of answer quality. Our GPT bot for Teams undergoes rigorous evaluation.
4. Deployment and monitoring: deployment to Azure Web App or AKS, logging configuration (Application Insights), hallucination monitoring. This RAG chatbot for Teams is production-ready.
5. Documentation and training: API specification, admin guide for Teams, operator training.

Our Microsoft Teams bot development process follows industry best practices, ensuring reliability and scalability.

Timeline and cost

A minimum MVP is delivered in 4-6 weeks. Typical investment for an MVP starts at $30,000–$50,000; a full solution with 3-5 system integrations ranges from $80,000 to $150,000. Cost is calculated individually after audit and depends on knowledge base size, business logic specifics, and security requirements.

Contact us for a project assessment — we guarantee a transparent work plan. With over 5 years of experience and 50+ deployed AI assistants, we deliver robust solutions. Request a consultation on AI assistant implementation today.

We specialize in Azure Bot Framework development for enterprise solutions. Our RAG bot for SharePoint ensures quick access to documents. LLM integration with Teams allows seamless interaction. Adaptive Cards in Teams provide rich interactive responses. This enterprise AI assistant improves productivity.

NLP Development: Text Classification, NER, Embeddings, and Information Extraction

We often receive a task: process 50,000 support tickets — currently all manual. Dataset — 3,000 labeled examples, 12 categories, imbalance: one category occupies 40% of the sample, three at 1-2% each. Baseline accuracy — 78%. Sounds decent until you look at recall for rare classes: 0.31, 0.44, 0.28. These classes — complaints and churn threats — are most important to the business.

This is a typical NLP development project. The problem is not the algorithm but that accuracy is the wrong metric. Our experience across 30+ projects shows: we start by analyzing business metrics and only then choose the model.

Why accuracy is not the right metric for rare classes?

Accuracy ignores imbalance. If the "churn" class appears in 2% of cases, the model can predict "all good" and get 98% accuracy — but the business loses clients. Solution: F1 macro (averaged over all classes) or weighted F1. For NER — strict entity F1 (exact matches only). We guarantee: after choosing the correct metric, model quality becomes measurable and predictable.

Text Classification: From BERT to Distillation

BERT-like models are the standard for classification. ruBERT-base or ruBERT-large from DeepPavlov for Russian. multilingual-e5-large — for multiple languages in one pipeline. XLM-RoBERTa-large — a strong multilingual backbone.

Fine-tuning for classification: add a classification head on top of the [CLS] token, train for 3-5 epochs with lr=2e-5, weight decay=0.01. For imbalance — weighted CrossEntropyLoss or focal loss with gamma=2.0. Contact us — we will show a code snippet.

Imbalance case study. Dataset — 3,000 examples, imbalance 1:20. Solution: class_weight via sklearn + CrossEntropyLoss. Additionally — augmentation of rare classes via backtranslation (ru→en→ru through MarianMT). Recall for rare classes rose from 0.31 to 0.67 with a slight drop in accuracy (76%→74%). Full NLP development end-to-end took 3 weeks.

Distillation for production. BERT-large gives F1 0.89, but inference on CPU — 180ms. Distillation into DistilBERT or ruBERT-tiny2 reduces latency to 25ms with F1 0.84. Export to ONNX Runtime provides an additional 1.5-2x speedup. DistilBERT achieves 7x lower latency than BERT-large with only a 5% drop in macro F1 – a typical production trade-off.

How to choose between BERT and LLM for your task?

For most classification and extraction tasks, BERT-sized models offer the best trade-off between cost and performance. Shift to LLMs only when the task demands generation, complex reasoning, or zero-shot generalization.

NER: Named Entity Recognition

NER — extracting persons, organizations, locations, dates, amounts, document numbers. For general categories (PER, ORG, LOC), pre-trained models work well. For specialized ones (medical terms, legal concepts) — fine-tuning is needed.

Data annotation. The main cost of an NER project. For a quality model — 500-2,000 labeled sentences per entity type. Tools: Label Studio (open source) or Prodigy (by spaCy creators). IOB2 format — standard.

Architecture. Token classification on top of BERT: each token gets a label (B-PER, I-PER, O). spaCy 3.x with transformer pipeline — a convenient production choice.

Nested entities. Standard IOB models cannot handle nested entities (organization inside an address). For such tasks — span-based NER: SpanBERT or SpERT. More complex but correct.

Post-processing is mandatory. The model predicts tokens — normalized entities are needed. Date — dateparser. Amounts — regex + validation. Names — deduplication via rapidfuzz. Included in our standard delivery.

Sentiment Analysis and Opinion Mining

Binary classification positive/negative works out of the box with BERT. Complexity — aspect-based sentiment analysis (ABSA): "the restaurant has good food but terrible service." For ABSA: aspect extraction (NER) + sentiment per aspect. Joint models BERT-for-ABSA — quality on Russian data is lower due to dataset scarcity. RuSentiment, SentiRuEval — main resources.

For production with simple positive/negative/neutral: distil models are enough. Three classes, balanced dataset, 2,000+ examples — F1 macro 0.82-0.87 in 1-2 days.

Text Summarization

Extractive summarization (select sentences) — TextRank or BM25 without training. Fast, no hallucinations. Good for long documents.

Abstractive (generates new text) — seq2seq: mT5, mBART, FRED-T5, ruT5-large. For production via LLM API (GPT-4, Claude) — often the best cost/quality/speed trade-off.

Embeddings: Vector Representations of Text

Embeddings are the foundation of semantic search, deduplication, clustering, RAG. Quality critically affects downstream tasks.

Models. E5-large-v2, BGE-M3, multilingual-e5-large — strong multilingual embedders. sentence-transformers/paraphrase-multilingual-mpnet-base-v2 — fast option. For Russian: ru-en-RoSBERTa (Skoltech) performs well on semantic textual similarity.

Embedding quality evaluation uses the MTEB benchmark as standard. But top results on MTEB don't guarantee success on a domain dataset — we build domain-specific eval.

Fine-tuning embeddings. If standard models don't give the required Recall@k — contrastive learning on domain pairs with MultipleNegativesRankingLoss. How to perform this for domain data:

1. Collect 500–2,000 semantically similar pairs from your domain.
2. Apply MultipleNegativesRankingLoss with a batch size of 32–64.
3. Train for 1–3 epochs using AdamW (lr=2e-5).
4. Evaluate Recall@k on a held-out domain test set.

This approach yields a 5–15% improvement in Recall@k in practice.

Dimensionality and storage. E5-large: 1024 dim, float32 — 4KB per vector. For 10M documents — 40GB. Quantization int8 reduces to 10GB. FAISS IVF_PQ — more compact but with losses. Included in our deployment recommendations.

Information Extraction

Structured extraction is a frequent task. Examples: key contract terms, technical characteristics, dates and amounts from invoices.

1. Regex + rule-based. For INN, OGRN, amounts, dates — more reliable than neural networks. No data required.
2. NER + post-processing. For variable formats.
3. LLM with structured output. GPT‑4 / Claude with JSON schema — for complex documents. Cost: minimal per document. For 10k+ documents/day — we calculate the economics.

We guarantee a hybrid: regex/NER for typical fields + LLM for edge cases. Our guarantee is backed by years of production experience and more than 30 projects.

Work Stages

Prototype on existing data — 1-3 weeks. Production system with CI/CD — 1.5–2.5 months. Cost is calculated individually — get a consultation for a project estimate.

What's Included

• Model and pipeline architecture documentation
• Access to the model via REST API (FastAPI + ONNX)
• Client team training (2-hour webinar + Q&A)
• Accuracy guarantee on the agreed test set
• Months of post-delivery support (bug fixes, adaptation to new data)

Our Experience

Years of NLP projects from classification to RAG systems. The team includes ML engineers experienced with Hugging Face, spaCy, LangChain, MLOps. We use vLLM, Kubeflow, Weights & Biases — a production stack, not toys. Contact us to evaluate your NLP project within two days — request a free consultation on your text processing pipeline.