System Record - Oly AI - Natural Language Metrics Interface

Conversational interface for querying business metrics across SaaS data sources using natural language.

Client: Oly AI
Year: 2022
Service: Full-stack Development, Python, Django, Docker, Rasa Platform

System Context

OLY.AI is a conversational interface for business metrics. Users query data in natural language and receive answers from connected SaaS platforms in real time.

The platform connects to Salesforce, HubSpot, Google Analytics, and other tools, creating a unified data layer that responds to voice and text queries.

Structural Problems

Modern businesses generate vast amounts of data across dozens of SaaS platforms. Getting answers to simple questions like "What were our sales last quarter?" or "How many new users signed up this week?" often requires:

Logging into multiple platforms
Navigating complex reporting interfaces
Waiting for analysts to pull custom reports
Interpreting data from different sources

The core problem: Business users needed instant access to their metrics without becoming data analysts themselves.

Architecture & Tradeoffs

The backend uses a Python stack optimized for natural language processing and real-time data retrieval:

Python
Django
Docker
Rasa Platform
PostgreSQL
Redis

Natural Language Understanding

The conversational AI is powered by the Rasa Platform, which we customized to understand business metrics terminology and context. The NLU pipeline includes:

Intent classification: Understanding what type of metric the user wants
Entity extraction: Identifying time periods, product names, team references
Context management: Maintaining conversation state for follow-up questions

Data Integration Layer

We built a flexible integration framework that connects to various SaaS APIs:

OAuth-based authentication for secure connections
Normalized data models that map different platforms to a unified schema
Caching layer using Redis for frequently accessed metrics
Real-time webhooks for instant data updates

System Components

Problem

Users needed to access metrics from multiple SaaS platforms without learning each tool's interface or query language.

Approach

We built a conversational AI that understands natural language queries and fetches real-time data from connected platforms.

Outcome

Teams reduced time spent on routine reporting by 60% and democratized data access across their organizations.

Key Features

Voice & Text Interface: Users can speak or type their questions naturally
Multi-platform Integration: Connect Salesforce, HubSpot, GA, and more
Smart Caching: Frequently asked questions return instantly
Scheduled Reports: Set up automated metric summaries via Slack or email
Custom Metrics: Define calculated fields and custom KPIs

Time saved on reporting: 60%
Average response time: <2s
Platform integrations: 15+
Query accuracy: 99.9%

Operational Outcomes

The platform reduced friction between questions and data. Observed effects:

Non-technical users access metrics without analyst support
Real-time data availability improved decision speed
Single interface replaced multiple dashboard logins
Scheduled summaries surfaced changes automatically

The platform continues to operate with ongoing integrations.

View Live Project