SpecOS

Website getspecos.com
SpecOS

SpecOS

Website getspecos.com

Role: Founder

  • Product Design & Branding
  • Full-Stack Development
  • Data Architecture & AI Integration

AI-powered product discovery for construction

SpecOS is a conversational and image-based product discovery platform built for construction. It takes product data from 385K+ SKUs and 1.2M structured variants and makes them searchable in seconds

Overview

Product data in construction is locked in catalogs and PDFs. Contractors waste hours chasing specs, while distributors lose sales when products can’t be found.

SpecOS transforms that workflow: ask in plain language or upload an image, and get structured results instantly.

Agentic RAG Model

Our system uses an Agentic Retrieval Augmented Generation (RAG) model. This is a multi-step process that uses specialized AI "agents" to deliver more accurate and context-aware results than a standard chatbot.

Svelte Flow
Press enter or space to select a node. You can then use the arrow keys to move the node around. Press delete to remove it and escape to cancel.
Press enter or space to select an edge. You can then press delete to remove it or escape to cancel.

Agentic RAG Model Diagram

Svelte Flow
Press enter or space to select a node. You can then use the arrow keys to move the node around. Press delete to remove it and escape to cancel.
Press enter or space to select an edge. You can then press delete to remove it or escape to cancel.

Data Model

The data model is designed for flexibility to accommodate complex and varied product catalogs without requiring schema changes.

Core Schema

A brand has many products.
A product has many product_variations.

Flexible Attributes

EAV Model To handle a wide range of product specifications, we use an Entity-Attribute-Value (EAV) model.

Product-specific attributes (e.g., "Height", "Voltage", "Material") are stored in the product_features table as key-value pairs.

Svelte Flow
Press enter or space to select a node. You can then use the arrow keys to move the node around. Press delete to remove it and escape to cancel.
Press enter or space to select an edge. You can then press delete to remove it or escape to cancel.

getProductAdvice Sequence

This diagram is best viewed on desktop

Hardcoded Sequences

Critical multi-step interactions are not left to the discretion of the AI. They are explicitly coded as non-negotiable sequences to ensure reliability and safety.

Example

The getProductAdvice Sequence When a user asks for product advice, the system executes a mandatory 3-message sequence. This is a hardcoded rule, not an AI decision.

Svelte Flow
Press enter or space to select a node. You can then use the arrow keys to move the node around. Press delete to remove it and escape to cancel.
Press enter or space to select an edge. You can then press delete to remove it or escape to cancel.

Reverse Image Search Architecture

This diagram is best viewed on desktop

Model Agnostic

SpecOS was designed to to model agnostic allowing customers to choose their preferred model without having to change the code.

Powered by OpenAI
Powered by Google Gemini
Powered by Claude

Project Details

Role: Founder

  • Product Design & Branding
  • Full-Stack Development
  • Data Architecture & AI Integration

Tech Stack

Frontend: SvelteKit, Tailwind CSS
Backend: Supabase, Pinecone Vector DB
AI: Google Gemini, OpenAI CLIP
Data: 400K+ SKUs, 1.9M structured variants
getspecos.com ↗
hero

© 2025 Ethan James Fox