SmartReconcile is a modular, AI-powered data reconciliation system designed to automate the process of tracking ingredients across purchase, production, and sales pipelines. It leverages multi-agent orchestration (LangChain, AutoGen, AutoGPT) and automation platforms (n8n, Airflow) to generate clean, consolidated reports with natural language summaries.

With advanced AI logic, customizable tooling, and seamless integration into low-code platforms, SmartReconcile empowers businesses to minimize manual reconciliation, detect anomalies in real-time, and improve operational visibility with PDF outputs.

LangChain agents use tools like Python, Pandas, and CSV loading to analyze tabular data with natural language prompts. They allow statements like:

“Which ingredient's production exceeded purchases?”

The create_pandas_dataframe_agent() from langchain_experimental enables querying merged DataFrames directly using LLMs.

Reference

AutoGen enables multi-agent workflows like:

• Data Agent: CSV ingestion
• Analysis Agent: Find anomalies
• Report Agent: Write insights

Its Studio GUI supports visual workflow design—making it a powerful no-code addition.

GitHub - microsoft/autogen

Designed to autonomously plan and complete tasks:

• Load → Analyze → Report
• Uses goal-directed loops with dynamic function calling

GitHub - Significant Gravitas AutoGPT

• reconcile.py: Core reconciliation script
• requirements.txt: Project dependencies
• Dummy Report 1.pdf and Dummy Report 2.pdf: Sample outputs
• Terttulia Purchase - Production - Sales Report.pdf: Annotated final report

git clone <this-repo>
cd smart-reconcile
pip install -r requirements.txt

.env file (optional):

OPENAI_API_KEY=your-key
USE_LLM_AGENT=true
PURCHASE_FILE=...
PRODUCTION_FILE=...
SALES_FILE=...

1. File Cache Check
2. CSV Cleaning & Normalization
   • Converts to kilograms
   • Handles missing or malformed units
3. Summary Table Construction
   • Merge purchase, production, sales
   • Fills missing SKUs with 0
4. Reconciliation Logic
   • Identifies mismatches
   • Groups by ingredient
5. PDF Report Generation
   • Multi-page summary
   • Overview + detailed per-SKU tables

• Total Purchased: 728.35 Kg
• Total Produced: 739.20 Kg
• Total Sold: 717.51 Kg
• Highest Purchased: Chicken Breast Boneless (168.3 Kg)
• Highest Sold: Chicken Breast Boneless (166.1 Kg)

(Full tables in the PDFs)

Check: Terttulia Purchase - Production - Sales Report.pdf

Covers:

• Yield calculations per dish
• Spoilage tracking
• Dish-wise breakdown (e.g. Butter Chicken Meal Box: 4 sold, 200g used each)

pandas
python-dotenv
langchain_experimental
langchain_openai
openai
matplotlib
fpdf

python reconcile.py

Generates:

• Reconciliation logic in pandas or via LangChain Agent
• Timestamped PDF in local folder

• Add cost columns to reconcile finance vs inventory
• Track yield % vs spoilage dynamically
• Integrate with BigQuery or Firebase for live data pulls

SmartReconcile is a powerful starting point for supply chain auditing, inventory management, or automated restaurant reporting. Combining Python with agentic AI provides:

• Automation of tedious joins
• Reliable, readable reporting
• Natural language narratives at scale

Happy Reconciling! 🚀