TaskWeaver (Microsoft) Integration for Analytical AI Agents

We've grown accustomed to an analyst spending a day on Excel, yet the task often boils down to a couple of queries. In one project, a three-person team manually prepared a monthly report — 40 hours each. TaskWeaver, a framework by Microsoft Research, turns an English description into executable Python code. We integrate such agents turnkey: connect sources, configure the sandbox, and teach the agent your business rules. As a result, a report that used to take a day is ready in half an hour, and the analyst shifts to meaningful tasks.

Problems We Solve

Errors during manual copying. Transferring formulas between Excel and reports introduces 2–3 errors per quarter. TaskWeaver generates code, eliminating the human factor: one formula once, not copied into 20 cells.

Non-reproducibility of reports. When an analyst leaves, their "recipe" is often lost. The agent stores the entire pipeline in code — it can be rerun with the same results a month later.

Slow processing of large datasets. Loading CSV via pandas in a notebook takes minutes, while the agent does it simply and quickly without a GUI.

How We Do It: Stack and Configs

We use the latest version of TaskWeaver (GitHub) with GPT-4o. The sandbox runs on Docker; allowed modules are limited to only those needed for analytics. For unstructured data, we set up RAG: the agent searches relevant documents in a vector database and uses them as context.

# Установка TaskWeaver
git clone https://github.com/microsoft/TaskWeaver.git
cd TaskWeaver
pip install -r requirements.txt

// project/taskweaver_config.json
{
  "llm.api_base": "https://api.openai.com/v1",
  "llm.api_key": "sk-...",
  "llm.model": "gpt-4o",
  "planner.example_base_path": "${AppBaseDir}/examples",
  "code_interpreter.use_local_uri": true,
  "code_interpreter.allowed_modules": ["pandas", "numpy", "matplotlib", "sklearn", "scipy"]
}

We write custom plugins for corporate databases. Example — db_query plugin for SQL queries:

from taskweaver.plugin import Plugin, register_plugin
import pandas as pd

@register_plugin
class DatabaseQueryPlugin(Plugin):
    def execute(self, query: str, database: str = "analytics") -> pd.DataFrame:
        conn = get_db_connection(database)
        return pd.read_sql(query, conn)

The plugin is registered in a yaml file, but we often configure it via code.

Practical Case: Financial Analysis in 25 Minutes

Problem: our client, a fintech company, spent 2 days monthly on a report: loading data from 3 sources, calculating 15 KPIs, building 8 charts, detecting anomalies.

Solution: we deployed a TaskWeaver agent that autonomously executes the entire cycle:

• Queries to PostgreSQL (revenue, costs)
• Loading Excel files (budget)
• KPI calculation via pandas
• Charting (matplotlib/plotly)
• Markdown report generation with insights

Results:

• Report preparation time: 2 days → 25 minutes automatic + 40 minutes review
• Calculation errors: 0 (previously 2–3 per quarter with manual copying)
• Analyst shifted to meaningful interpretation instead of routine

Why TaskWeaver is Better than Standard Code Interpreter?

ChatGPT Code Interpreter breaks context with each run. TaskWeaver maintains session state: you can do step 1, pause, view intermediate results, then command "repeat analysis only for region A". This gives 5 times more control over multi-step analysis.

from taskweaver.app.app import TaskWeaverApp

app = TaskWeaverApp(app_dir="./project")
session = app.get_session()

# Шаг 1: загрузка и очистка
session.chat("Загрузи данные продаж из БД за прошедший год, удали дубликаты")
# Шаг 2: анализ сезонности
session.chat("Проведи STL-декомпозицию")
# Шаг 3: прогноз
session.chat("Построй прогноз на следующий квартал с помощью Prophet с backtesting")
# Шаг 4: отчёт
result = session.chat("Сформируй markdown-отчёт с графиками")
print(result.post_list[-1].get_text())

What LLM Models Can Be Used with TaskWeaver?

Any OpenAI (GPT-4o, GPT-4), Claude, LLaMA 3, Mistral are supported. We typically use GPT-4o for its high code generation accuracy. The LLM can be changed in the config without reinstalling the framework. For tasks with high latency, Mistral is suitable — it is faster but slightly less accurate. The choice of model depends on your requirements for speed and accuracy.

How TaskWeaver Handles Multi-step Tasks?

The planner breaks down the task into sub-tasks, turns each into code, executes, and analyzes the result. If an error occurs at step 2, the agent reformulates the code automatically. We ensure accuracy through chain-of-thought prompting: the agent writes a plan before code, then executes.

Implementation Process

1. Audit sources and business logic. Determine required plugins, allowed modules, execution frequency.
2. Design the agent. Select LLM, configure sandbox, write custom plugins.
3. Development and testing on real data. 3–10 iterations.
4. Production deployment. Docker, scheduler (cron/airflow), monitoring.
5. Team training. 2 workshops, documentation.

Scope of Work

Timelines and Team Experience

• Basic setup: 2–3 days.
• Full cycle with plugins and testing: 1–2 weeks.
• We have deployed TaskWeaver for 10+ companies in finance, retail, and logistics. We have been working with LLMs since the first production-ready models — we know all the nuances of prompt engineering (few-shot, chain-of-thought).

Estimate your project. Contact us, describe your task — we will send a demo session of your agent in 2 days. We guarantee the sandbox is secure and code does not leak. Contact us for a consultation and precise timeline estimate.

LLM Development: Fine-Tuning, RAG, Agents, and Production Deployment

Using GPT‑4 or Claude 3.5 Sonnet through a public API is not a solution — it's just a tool. When the requirement is to "make it like ChatGPT, but on our data," there is a real engineering challenge behind it: from prompt engineering to training a 70B model on your own infrastructure. End-to-end LLM solution development is a complex stack, and we have been doing it for over 5 years. During this time, we have completed over 20 projects in generative AI: from RAG systems for legal departments to custom support agents. Where exactly your task falls depends on data, latency requirements, budget, and how critical confidentiality is.

A typical situation: the client has already tried ChatGPT, but results are unstable — sometimes accurate, sometimes hallucinating. Or they need integration into a corporate portal while complying with security policies. Let's break down each layer of the stack in detail — from RAG to production deployment.

Why Do RAG Systems Break and How to Fix It?

RAG (Retrieval-Augmented Generation) looks simple: find relevant documents, put them in context, get an answer. In practice, it fails in several places.

Chunking without overlap. Classic mistake: chunk_size=512, overlap=0. If the answer lies across two chunks, retrieval won't find either with sufficient confidence. Solution: overlap 15–25% of chunk_size, or better yet, sentence-aware splitting with spaCy or NLTK instead of naive character splitting.

Poor embedder. text-embedding-ada-002 is good for general use, but on legal or medical texts, specialized models like E5-large-v2, BGE-M3, or fine-tuned sentence-transformers on domain data outperform it. Recall@5 differences can be 15–25%.

No re-ranking. Vector search optimizes for speed, not relevance. A cross-encoder re-ranker (ms-marco-MiniLM-L-6-v2, bge-reranker-large) after initial retrieval improves top-3 accuracy with acceptable latency (+50–150ms). This is often more impactful than improving the embedding model.

Hybrid search. Dense vectors alone work poorly on exact queries: names, SKUs, codes. BM25 (sparse) finds exact matches but misses semantics. Hybrid via RRF (Reciprocal Rank Fusion) is the optimal compromise. Qdrant, Weaviate, and pgvector 0.7+ support hybrid search natively.

When to Fine-Tune Instead of Prompt Engineering?

Prompt engineering solves ~70% of LLM adaptation tasks for a domain. The remaining 30% require fine-tuning. Three indicators: the model ignores a specific output format even with detailed prompting; the task requires deep knowledge of specialized vocabulary (medicine, law); you need to significantly reduce token costs by replacing a large model with a smaller specialized one.

LoRA and QLoRA are the standard for SFT. LoRA adds trainable low-rank matrices to attention layers. A typical configuration for Llama-3 8B: r=64, lora_alpha=128, target_modules=["q_proj","v_proj","k_proj","o_proj"] yields ~0.8% trainable parameters, training on one A100 40GB. QLoRA adds 4-bit quantization (NF4) and allows fine-tuning 70B models on two A100 40GB, though speed drops by half compared to bf16.

DPO instead of RLHF. Direct Preference Optimization requires only (chosen, rejected) pairs, not scalar reward signals. DPOTrainer from the trl library (Hugging Face) implements it in a few dozen lines.

Common mistake. A dataset of 500 examples, 5 epochs, validation loss 0.8 — seems fine. But on test, the model degrades on general instructions. Cause: catastrophic forgetting. Solution: add 10–20% general instruction-following examples (Alpaca, FLAN) to the training set to preserve original capabilities.

How to Choose a Base Model: 8B or 70B?

For most tasks, fine-tuning an 8B model is sufficient. 70B is needed when deep reasoning is required or the 8B baseline does not reach the required quality even after fine-tuning. Inference cost for Llama-3 8B via vLLM on A100 is efficient; the exact cost depends on volume.

What Does PagedAttention Bring to Production?

vLLM is the first choice for serving open-source models. PagedAttention is the key technical innovation: KV-cache is managed like virtual memory in an OS, without fragmentation. This yields 2–4x higher throughput compared to naive HuggingFace Transformers inference. The vLLM documentation confirms that continuous batching and PagedAttention are the standard for high-load LLM services.

Typical numbers on A100 80GB for Llama-3 8B (bf16): 400–600 req/s, P50 latency 200–400ms, P99 latency 600–900ms at concurrency 64. For 70B on two A100 with tensor parallelism: 80–120 req/s, P99 latency 1.5–2.5s. AWQ or GPTQ quantization reduces memory consumption by 2x with quality loss within 1–3%.

Multi-Agent Systems

Agents are LLMs with access to tools: search, code execution, API calls, database interaction. Common patterns:

• ReAct (Reason + Act): the model reasons → chooses a tool → observes the result → reasons again. LangChain and LlamaIndex implement it out of the box.
• Multi-agent orchestration: multiple specialized agents with a coordinator on top. Example: coordinator → researcher (search + summarization) → coder (code generation and execution) → critic (verification). Tools: AutoGen (Microsoft), CrewAI, custom implementation on LangGraph.

In production, agent systems are non-deterministic. Essential: guardrails, step limits, logging of each step, human-in-the-loop for critical actions.

How We Work: Stages, Timeline, Deliverables

What Is Included

We deliver:

• Technical documentation (model card, configs, deployment instructions)
• Access to infrastructure (code repository, trained weights)
• 1 month of post-deployment support (consultations, bug fixes)
• Customer team training (2–3 sessions on system operation)

Timeline: basic RAG prototype — 1–2 weeks. Fine-tuning with customer data — 3–6 weeks (including data preparation). Production system with monitoring and retraining — 2–4 months. Cost is calculated individually based on data volume, model complexity, and infrastructure requirements.

We guarantee the quality of the final model with performance benchmarks and ongoing monitoring. Our engineers have hands‑on experience with dozens of production LLM systems.

Want to evaluate your project? Leave a request — we will prepare a preliminary summary within 1–2 business days. Or get a consultation on choosing the approach: RAG, fine-tuning, or hybrid — we will tell you what works best for you. Contact us to discuss your LLM development needs. Schedule a free consultation today.