# NOVA: Building Agentic Workflows for Structured Data Intelligence at Nexla

**Date**: _April 10, 2025_

Speaker:

Noel Nebu Panicker

AI Engineer at Nexla

## Watch the Talk

NOVA: Building Agentic Workflows for Structured Data Intelligence at Nexla - YouTube

Tap to unmute

[NOVA: Building Agentic Workflows for Structured Data Intelligence at Nexla](https://www.youtube.com/watch?v=djUFEs3fM5s) [AG2](https://www.youtube.com/channel/UCp8MDiqezK6w80-7x3A_9-Q)

AG2993 subscribers

## Talk Recap

In this talk, Noel introduced **NOVA**, Nexla's AI-powered co-pilot designed to address one of the toughest challenges in enterprise environments: **integrating scattered data across diverse tools, systems, and formats**.

Built on an **agentic foundation powered by AG2**, NOVA simplifies how both technical and non-technical users can build and manage complex data workflows.

## Solving Enterprise Data Complexity with NOVA

A central concept behind NOVA is the need to **ground AI outputs in real, production-grade data**. Noel explained how Nexla’s **metadata intelligence layer** powers NOVA with rich contextual awareness — including field types, sample values, and data relationships — reducing hallucinations and enabling **safe, data-aware code generation** and seamless API integrations.

## Intelligent Data Mapping & Extraction

NOVA also functions as an **intelligent translator**, capable of:

- Mapping messy, inconsistent inputs into structured, enterprise-standard models.
- Accelerating **data onboarding and transformation processes**.
- Extracting structured information from unstructured sources like **emails and scanned documents** using models such as **Pixtral** and **GPT-4**.

## How NOVA Operates

Noel shared a breakdown of NOVA’s **agentic workflow architecture**:

- **Initializer Agent**: Sets up the transformation requests.
- **Planner Agent**: Converts user intent into a series of actionable tasks.
- **Query Interpreter Agent**: Maps these tasks to specific data transformations.
- **Structured Extraction Workflows**: Extract structured data from unstructured inputs.

The system is **metadata-driven**, supports **real-time updates through Server-Sent Events (SSE)**, and is designed for **explainability, control, and traceability**.

## Real-World Impact

NOVA has already demonstrated significant value:

- 📈 **10% increase** in pipeline creation rates since launch.
- ⏳ **Reduction in partner onboarding time** from 6 weeks to 2 days.
- 🔒 Enabled the creation of **secure, governance-friendly private data marketplaces** for non-technical users.

These outcomes point toward a promising future for **AI-native, data-driven enterprise operations**.

## Discover More

Curious about how all of this comes together? The full user story is captured in the [Unlocking the Power of Agentic Workflows at Nexla with AG2](https://docs.ag2.ai/0.10.2/docs/user-stories/2025-02-11-NOVA/nova) section of this document. Be sure to check it out.