Invisible Ranks 61st Fastest-Growing Company on the 2024 Deloitte Technology Fast 500™. Read More.
In 2020, Boosted.ai expanded its AI-powered financial analysis platform, Boosted Insights, by developing an AI portfolio assistant for asset managers using a large language model (LLM). The assistant processes data from 150,000 non-traditional sources to deliver macro insights and trend analysis on over 60,000 stocks across global equity markets overnight.
While Boosted.ai's customers were responding well to the initial product, they wanted more precise insights, broader industry coverage, and faster speed. To meet these demands, Boosted.ai would need to process up to 10 times more data while reducing processing time by 99.99%.
Between 2023 and 2024, Microsoft, Google, Meta, and Salesforce began building AI products with small models trained on specific concepts and processes for defined use cases. These Small Language Models (SLMs) are less than 0.003% the size of typical LLMs and drastically reduce the time, energy, and costs of developing AI products. Boosted.ai realized that powering their AI investment assistant with an SLM could help them achieve their goals and reached out to AWS for support.
With AWS support, Boosted.ai saw promising results fine-tuning an open-source SLM. However, without more advanced ground truth training data, the SLM couldn’t meet their standards. They attempted other solutions with limited success, and AWS connected Boosted.ai with Invisible for a solution.
Invisible’s team collaborated with Boosted.ai to produce precise data for the SLM to learn and apply financial expertise accurately. This extended beyond data labeling to include data guiding the model on when to apply financial concepts and execute multi-step reasoning.
Invisible's Trainer teams checked for common-sense errors and applied expert-level QA. The work was complex, with tasks taking up to two hours each.
By the third data batch, Boosted.ai felt “unlocked.” Their product and tech teams could stop focusing on data quality and start running experiments to optimize customer value and ROI. A key benefit of working with Invisible, Boosted noted, was its meticulous metadata and documentation, which led to new insights.