← Back

May 2026

ConvoFill

Agentic conversational data extraction system for LetSale

View on GitHub

Overview

ConvoFill is a capstone machine learning project built for the LetSale marketplace app. It replaces traditional form-based item listing with an agentic conversational interface that extracts structured data from natural dialogue.

Features

Multi-Agent Orchestration

A coordinated system of specialized agents handles intent detection, context gathering, extraction, and validation.

Structured Data Extraction

Converts unstructured conversation into strict JSON schemas using Pydantic validation.

Real-Time Inference

Uses Groq for low-latency conversational responses and fast extraction cycles.

Schema Validation Layer

Ensures extracted outputs conform to marketplace requirements before persistence.

Technologies

• Python
• Swift (iOS)
• Flask API
• Groq API
• LangChain
• Pydantic
• AWS SageMaker
• Railway deployment

Impact

ConvoFill shows how LLMs can sit between users and apps to replace fixed forms with a more natural input flow, making it easier to submit structured data.