AI Agent for Financial Analysis Development

How an AI Agent Transforms Financial Analysis?

We develop AI agents that automate financial analysis: from data collection to report generation with recommendations. Our experience — over 10 projects in the financial sector. The combination of Text-to-SQL for data operations, Code Interpreter for calculations, and LLM for interpretation makes the agent capable of answering complex analytical questions without a finance professional's involvement. By reaching out to us, you get a ready-made solution integrated with your ERP. We guarantee calculation accuracy according to agreed metrics.

What Problems Does a Financial AI Agent Solve?

Financial analysts spend up to 40% of their time on manual data collection and formula verification. Errors in EBITDA or ROE calculations can cost millions. Our agent uses Text-to-SQL for direct database access and Code Interpreter for precise computations — eliminating the human factor. Additionally, the agent automatically detects anomalies using statistical methods (z-score), allowing issues to be identified before they affect reporting.

Why Is an AI Agent More Accurate Than Manual Analysis?

In manual analysis, calculation errors depend on employee qualification. An AI agent uses predefined formulas and verifies each step through tools. For example, during plan-fact analysis, the agent executes SQL queries, calculates metrics by strict formulas, and decomposes variances into price and volume effects. Result: calculation error approaches zero, and preparation time is reduced by 4–8 times.

How We Design and Implement the Agent

We use GPT-4o with tool use functions, Hugging Face for embeddings, pgvector for semantic search over financial documentation. For deployment — vLLM or SageMaker. Each agent undergoes calculation verification with the finance department. At the core is a system prompt with factor analysis methodologies (price/volume effect) and confidence intervals.

What Is Included in Turnkey AI Agent Development?

• Integration with data sources (PostgreSQL, 1C, ERP)
• Custom toolset (SQL functions, metric calculation)
• System prompt and few-shot example configuration
• Interaction interface (web chat or API)
• Documentation and training for the financial team
• Guarantee on calculation accuracy per agreed metrics

Development Stages

Practical Case: Plan-Fact Analysis for a Manufacturing Company (Our Experience)

Task: Monthly plan-fact P&L analysis by product lines and regions. Previously took 2 days for a financial analyst.

Solution: Our agent with tools for PostgreSQL queries, metric calculation, and model building.

Example interaction:

Query: "Analyze budget execution for revenue in March. Identify deviation causes."

The agent executed a sequence of calls: SQL query, metric calculation, variance decomposition, waterfall generation. Result: "Total deviation -8.3M rubles (-4.2%). Main factors: decline in product A sales volume (-5.1M, volume effect), partially offset by price increase on product B (+2.1M, price effect). Central Federal District region — only with overperformance (+1.8M), Urals — largest shortfall (-6.2M)..."

Implementation Results:

• Report preparation time reduced from 2 days to 3.5 hours
• Indicator coverage identical to manual analysis
• Interpretation quality (CFO evaluation) — 4.1 out of 5.0
• Savings on analyst overtime — up to 2.5M rubles per year

Comparison: Manual Analysis vs. AI Agent

from openai import OpenAI
from pydantic import BaseModel
from typing import Literal, Optional
import pandas as pd
import json

client = OpenAI()

financial_tools = [
    {
        "type": "function",
        "function": {
            "name": "query_financial_database",
            "description": "Query financial database (revenue, expenses, budget, actual)",
            "parameters": {
                "type": "object",
                "properties": {
                    "sql_query": {"type": "string"},
                    "description": {"type": "string"},
                },
                "required": ["sql_query"]
            }
        }
    },
    {
        "type": "function",
        "function": {
            "name": "calculate_financial_metrics",
            "description": "Calculate financial indicators",
            "parameters": {
                "type": "object",
                "properties": {
                    "metric": {
                        "type": "string",
                        "enum": ["EBITDA", "ROE", "ROA", "ROIC", "NPV", "IRR", "payback_period",
                                 "gross_margin", "operating_margin", "net_margin", "current_ratio",
                                 "debt_to_equity", "working_capital"]
                    },
                    "input_data": {"type": "object"},
                },
                "required": ["metric", "input_data"]
            }
        }
    },
    {
        "type": "function",
        "function": {
            "name": "build_financial_model",
            "description": "Build financial model (DCF, budget, P&L forecast)",
            "parameters": {
                "type": "object",
                "properties": {
                    "model_type": {"type": "string", "enum": ["dcf", "budget_variance", "pnl_forecast"]},
                    "parameters": {"type": "object"},
                },
                "required": ["model_type", "parameters"]
            }
        }
    },
    {
        "type": "function",
        "function": {
            "name": "generate_financial_report",
            "description": "Generate financial report",
            "parameters": {
                "type": "object",
                "properties": {
                    "report_type": {"type": "string"},
                    "period": {"type": "string"},
                    "data": {"type": "object"},
                },
                "required": ["report_type", "period"]
            }
        }
    },
]

def calculate_financial_metrics(metric: str, input_data: dict) -> str:
    """Accurate calculation of financial metrics"""

    calculators = {
        "EBITDA": lambda d: d["revenue"] - d["cogs"] - d["opex"] + d.get("da", 0),
        "gross_margin": lambda d: (d["revenue"] - d["cogs"]) / d["revenue"] * 100,
        "operating_margin": lambda d: d["ebit"] / d["revenue"] * 100,
        "ROE": lambda d: d["net_income"] / d["equity"] * 100,
        "ROA": lambda d: d["net_income"] / d["total_assets"] * 100,
        "current_ratio": lambda d: d["current_assets"] / d["current_liabilities"],
        "debt_to_equity": lambda d: d["total_debt"] / d["equity"],
    }

    calculator = calculators.get(metric)
    if not calculator:
        return f"Metric {metric} not implemented"

    try:
        result = calculator(input_data)
        return json.dumps({
            "metric": metric,
            "result": round(result, 4),
            "unit": "%" if metric in ["gross_margin", "operating_margin", "ROE", "ROA"] else "x",
        })
    except KeyError as e:
        return f"Missing required field: {e}"
    except ZeroDivisionError:
        return "Division by zero: check denominator values"

FINANCIAL_ANALYST_PROMPT = """You are a CFO-level financial analyst.

Your tasks:
1. Analyze financial data accurately and methodologically correctly
2. Use tools for calculations — never calculate in your head
3. For plan/fact variances — identify causes (price effect, volume effect, mix)
4. Give specific recommendations, not abstract observations
5. Point out anomalies and potential risks

Methodology:
- When analyzing P&L, break down variances into price and volume effects
- When assessing efficiency, compare with standard industry benchmarks
- For forecasts, indicate confidence interval and key assumptions"""

def financial_analysis_agent(question: str, context_data: dict = None) -> str:
    messages = [
        {"role": "system", "content": FINANCIAL_ANALYST_PROMPT},
        {"role": "user", "content": question},
    ]

    if context_data:
        messages.insert(1, {
            "role": "system",
            "content": f"Data context:\n{json.dumps(context_data, ensure_ascii=False, indent=2)}"
        })

    # Agent loop
    for _ in range(8):
        response = client.chat.completions.create(
            model="gpt-4o",
            messages=messages,
            tools=financial_tools,
        )

        msg = response.choices[0].message
        messages.append(msg)

        if not msg.tool_calls:
            return msg.content

        for tool_call in msg.tool_calls:
            tool_name = tool_call.function.name
            tool_args = json.loads(tool_call.function.arguments)
            result = execute_financial_tool(tool_name, tool_args)
            messages.append({
                "role": "tool",
                "tool_call_id": tool_call.id,
                "content": result,
            })

def detect_anomalies_in_data(financial_data: pd.DataFrame) -> list[dict]:
    """Statistical anomaly detection before passing to LLM"""

    anomalies = []

    for column in financial_data.select_dtypes(include="number").columns:
        mean = financial_data[column].mean()
        std = financial_data[column].std()
        z_scores = (financial_data[column] - mean) / std

        outliers = financial_data[abs(z_scores) > 2.5]
        if not outliers.empty:
            for idx, row in outliers.iterrows():
                anomalies.append({
                    "column": column,
                    "value": row[column],
                    "z_score": round(z_scores[idx], 2),
                    "period": str(idx),
                })

    return anomalies

Timeline and Cost

• Design: 1 week
• Development: 2–3 weeks
• Integration and testing: 2–3 weeks
• Verification with finance team: 2 weeks
• Total: 7–10 weeks. Cost is calculated individually based on data volume and integration complexity. Contact us for a project evaluation.

Order AI agent development for your tasks. Get a consultation — we will assess your project and offer the optimal solution.

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.