Floworks, a Bengaluru and San Francisco-based Generative AI company recently announced the launch of Mixture of Agents (MoA) ThorV2 and its first application, Alisha, the world’s first fully autonomous AI-powered Sales Development Representative (SDR).
Floworks’ innovative technology addresses the limitations of the current Large Language Models (LLMs) by offering superior accuracy, reliability, and cost-efficiency.
The Indian AI industry is experiencing explosive growth, driven by a surge in investment and a vast pool of talented engineers. However, existing Large Language Models (LLMs) like GPT-4o, GPT4 Turbo, and Claude-3 Opus struggle with reliability, often generating inconsistent results for the same prompt. This inconsistency hinders their real-world application, particularly in critical business tasks, where a single miss can cost millions of dollars.
Floworks, founded by Sarthak Shrivastava and Sudipta Biswas, aims to bridge this gap with its revolutionary ThorV2 engine. Unlike other LLMs, ThorV2 is specifically designed for high reliability, ensuring consistent and accurate outputs, making it ideal for powering AI employees who can perform real-world tasks effectively.
“We are incredibly excited to introduce ThorV2 and Alisha,” said Sarthak Shrivastava, Co-founder of Floworks. “Our vision is to empower Indian entrepreneurs and businesses globally with the tools they need to thrive. Floworks’ AI employees are poised to revolutionize business operations, particularly for growing startups and small and medium enterprises (SMEs) in India.”
ThorV2: The Engine Powering AI Employees
Floworks’ ThorV2 engine is an LLM architecture built from the ground up to address the shortcomings of existing LLMs like GPT4o, GPT4 turbo, and Claude 3 Opus regarding tool usage. Here’s how ThorV2 stands out:
-
Prioritizing Reliability: ThorV2 overcomes the challenge of inconsistency by prioritizing reliability, which is key to building any fully autonomous AI agents or AI employees. This ensures that AI employees perform tasks consistently and accurately, even with repeated prompts.
-
Superior Performance: ThorV2 outperforms current state-of-the-art models in four key categories:
-
Accuracy: ThorV2 is the most accurate tool-using model in the world right now, with 90.1% accuracy, which is a big jump compared to accuracies of Claude-3 Opus (78.9%), GPT-4o ( 53.5%) and GPT-4 turbo ( 48.6%). Unlike traditional LLMs prone to “hallucination” (generating irrelevant outputs), ThorV2 delivers highly accurate results, consistently understanding prompts and generating relevant responses.
-
Latency: ThorV2 has an average latency of 2.29 seconds for tool usage, 27.5% faster than GPT-4o (the fastest state-of-the-art model in the world). Speed is crucial in business. ThorV2 boasts exceptional efficiency, processing requests and delivering results significantly faster than competing models.
-
Cost: In comparison to others, ThorV2 costs $1.60 for 1000 queries which is 175% cheaper than GPT-4o (the cheapest state-of-the-art model in the world). Traditional LLMs can be resource-intensive for tool usage. ThorV2 is remarkably cost-effective, requiring a fraction of the resources needed to operate other LLMs, making it a budget-friendly solution.
-
Reliability: ThorV2 is the most reliable model with near zero hallucination and reliability for function calling of 99.5%, with other state-of-the-art models like Claude-3 Opus (59.7%), GPT-4o (83.9%) and GPT-4 turbo (86.6%) lagging significantly.
-
Introducing Alisha: The AI SDR
Alisha is the world’s first fully autonomous AI employee built on the foundation of ThorV2. Alisha automates the entire lead generation process, from prospecting and qualification to scheduling meetings, allowing human SDRs to focus on closing deals. Alisha can:
-
Identify and qualify leads, saving human SDRs valuable time and effort.
-
Craft personalized outreach emails, ensuring higher engagement rates.
-
Dynamically trains ThorV2 with product and company information to instantly reply to prospects’ queries and questions over emails- Coordinates with the prospect like a human and schedules sales meetings
“Alisha empowers sales teams to achieve unprecedented levels of productivity for sales teams, our customers are already seeing 4x more sales calls being booked per SDR. It’s almost like magic. You wake up in the morning to find 10 meetings booked in your calendar!” said Sudipta Biswas, Co-founder of Floworks. “By automating prospecting, the most time-consuming aspect of the SDR role, Alisha frees up valuable time for SDRs to connect with qualified leads and close more deals.”
Customers are using Alisha to generate thousands of sales calls with qualified leads per week, resulting in thousands of dollars in additional revenue. With Alisha by their side, SDRs can prospect ten times more leads each day, significantly boosting their sales performance and efficiency.
The Future of Work: A Workforce Augmented by AI
Floworks is dedicated to developing a suite of AI-powered employees that empower businesses to thrive in the AI age. In addition to Alisha, Floworks is developing a range of AI employee solutions, including AI RevOps professionals, AI project managers, and AI Executive Assistants. These solutions are designed to streamline various business functions and empower entrepreneurs to become “one-person unicorns” – highly successful businesses run by a single individual with the help of a team of AI employees.
“Engines like Floworks’ ThorV2 help achieve enterprise-grade reliability and performance with tool use while working with highly probabilistic large-language models. It allows a much-needed upgrade for generative AI to break into serious business applications,” adds Shival Gupta, Director of Engineering, and Nirav Bhan, Head of AI at Floworks and authors of the ThorV2 architecture.
Both Shival and Nirav come from storied backgrounds. Shival previously founded and exited multiple startups right from his college days and has been an engineering leader at multiple large organizations. Shival is also a renowned author of horror novels. Nirav was a child prodigy and a gold medalist at the National Physics Olympiad He later went through the Massachusetts Institute of Technology (MIT) to pursue research in Machine Intelligence before joining Floworks.
