Case Studies

End-to-end AI solutions, trusted by leading companies.

Request a Consult
Text
Danny Moldovan
,
Moldovan
Co-Founder at Change.org

FractionalAI has been invaluable in leveling up our internal gen AI automations. Their team took the time to thoroughly understand our unique issues and requirements, then applied their AI expertise to our content moderation workflows, significantly boosting our system’s accuracy and efficiency. They packaged their custom built solution in a flexible, API-driven component, enabling seamless deployment. I just hired them again, this time to build automations to support my PR team.

Text
Ian Meyers
,
Co-Founder at Sincera

We had a tough AI problem that we wanted to solve quickly and systematically. Fractional AI was an incredible thought partner that embedded seamlessly with our team.

Not only were we able to solve our immediate challenge our team leveled up our own AI skills for the future.

LinkedIn
Michel Tricot
,
CEO & Co-Founder at Airbyte

AI Assist in the builder was a huge milestone for our 1.0 launch! As many of you know, going from Prototype to Live when dealing with LLMs is a challenge for many companies.

On our side we worked alongside Fractional AI to get this amazing product live :)

X (Twitter)
Nathaniel Whittemore
,
CEO at Superintelligent

Have been collaborating with these guys on something I get to share soon and HIGHLY recommend them

End-to-end AI solutions,
trusted by leading companies.

Fractional AI Case Studies

Case study

Zapier: Reducing Hallucinations by Over 80%

Fractional AI partnered with Zapier to take their AI API integration builder to the next level: defining a robust suite of LLM evaluations and driving down hallucinations by over 80%.

Problem

Building and maintaining integrations for over 7,000 apps is no small task, so Zapier developed a system to automatically generate OpenAPI specs.

The problem? Performance was a black box, leaving the team without data to guide improvements or assess the impact of changes.

Solution

Fractional AI teamed up with Zapier to define robust LLM evaluations and use this data to iteratively improve the system's performance.

Models used: GPT-4o, Claude Sonnet-3.5
Tools used: Braintrust

Impact

The Zapier team saw dramatic improvements across key metrics -- with hallucinations decreasing by over 80%.

This more reliable system means more integrations with Zapier and more time saved for engineers building integrations.

Case study

Change.org: Automating Content Moderation

Change.org, the world’s largest platform for social change, enables anyone to start campaigns and mobilize support. With thousands of campaigns launched daily, not all adhere to Change.org's trust and safety terms. Fractional AI partnered with Change.org to automate content moderation, so the team can spend less time reviewing content and more time supporting changemakers.

Problem

The Change.org back-office team built an impressive system in Google Sheets that used LLMs to proactively flag half of violations, but finding the other half was still a big drain on time and the site experience.

Solution

Fractional AI built an AI-powered content moderation system to automatically review content that violates Change.org’s guidelines.

Models used: GPT 4o, we fine-tuned GPT 3.5 using OpenPipe
Tools used: Langchain, Langsmith

Impact

The new system catches 77% of violations while halving the proportion of false positives.

Case study

Airbyte: Taking API Integrations from Hours to Minutes

Airbyte is the leading open-source data integration engine that helps you consolidate your data in your warehouses, lakes, and databases. Airbyte’s users spend a lot of time building API integrations – a complicated and time-intensive process, so Airbyte hired Fractional AI to develop an AI-powered connector builder, producing API integrations in minutes, not hours.

Problem

Building API integrations is tedious and complicated.

Engineers have to navigate lengthy API docs, extract relevant details, and manually configure complicated connectors in Airbyte’s tools – all of which takes time away from other engineering projects.

Solution

Users input the API docs URL, and the AI Connector Builder, built by Fractional AI, crawls API docs and configures an Airbyte connector.

Models used: GPT 4o
Tools used: OpenAI’s SDK, Langsmith, OpenAI's built-in vector store

Impact

You can now build an API integration in less than 5 minutes with the help of AI.

Case study

Superintelligent: Personalizing Learning with RAG

Superintelligent is the learning platform for AI, offering tutorials on AI tools that are hands-on, practical, and easy to follow. When debuting their platform, Superintelligent imagined a fully personalized, AI-powered user experience, but where do you start with a project that big? Fractional AI partnered with Superintelligent to build an AI chatbot as the first step towards that vision.

Problem

Learners needed a way to quickly navigate the platform’s hundreds of tutorials and tool recommendations to find relevant AI tools for their specific needs.

Solution

Fractional AI developed an AI chatbot using RAG to recommend specific tools to learners.

Models used: GPT 4, GPT 3.5 Turbo
Tools used: Langchain, Langsmith, Pinecone, FastAPI

Impact

The chatbot has given hundreds of recommendations - eliminating the need for extensive searches, so users can spend less time searching and more time learning.

Case study

Sincera: Unlocking the value of unstructured data

Sincera, a technology platform leveraging metadata to enhance adtech solutions, faced a major challenge: a vast stream of unstructured data. Fractional AI partnered with Sincera to build an AI system to normalize and classify this data, unlocking its potential for value creation.

Problem

Sincera’s lean team receives millions of unstructured records each month, detailing thousands of distinct products and services (in CSVs!). The lack of consistency in syntax and structure makes it impossible to use the data without lengthy and tedious standardization efforts.

Solution

Fractional AI developed an AI categorization workflow to map unstructured, arbitrary product data into a clear product taxonomy in real-time. 

Models used: GPT-4o-mini
Tools used:
Braintrust

Impact

Fractional AI's system enables Sincera to confidently process this messy data in an automated pipeline without costly human intervention, making the data far more usable than would otherwise be possible.