UHG
Search
Close this search box.

AIM Exclusive: YC-backed Indian Startup Claims its AI Agent is Better than OpenAI’s GPT-4o

ThorV2 –developed by FloWorks—costs just $1.60 per 1000 queries, making it 175% cheaper than GPT-4o.

Share

floworks

Are AI agents the next big thing? Bengaluru and San Francisco-based startup Floworks definitely think so. The startup, funded by Y Combinator in the Winter 23 cohort, is building AI agents that can handle end-to-end sales functions in an organisation.

In an exclusive interaction with AIM, Floworks co-founders Sudipta Biswas and Sarthak Shrivastava said that the startup is building what they call an ‘AI employee’. 

Large language models (LLMs) today can’t do certain things despite their impressive knowledge base, like sending an email. The startup, which raised $1.5 million in seed funding earlier this year, has developed Alisha, which they call the world’s first fully autonomous AI-powered sales development representative (SDR).

Alisha can automate the entire lead generation process, from prospecting and qualification to scheduling meetings and sending emails, allowing human sales representatives to focus on more concrete tasks.

Alisha Achieves 100% Reliability 

The startup claims that with Alisha, users will generate 10x more leads each day. The AI assistant, which is powered by Thorv2, can interact with external tools such as emails, CRMs, Google Search, and Calendars. It can also parse through one’s email for leads and update the CRM accordingly.

“On any given day, we probably use 10 to 15 different software or tools. But a standard language model like ChatGPT cannot use these tools, and this is where we come in. 

“Our model, which we are internally calling ThorV2, is the most accurate and the most reliable model out there in the world when it comes to using external tools right now,” Biswas said.

Alisha is a product that comes out of the box and can be set up in under five minutes. However, the startup helps fine-tune Alisha to the particular business’s use case by training Alisha with that particular company’s data. 

“Alisha, our SDR, trains itself using this data, integrating workflows and processes specific to each customer. Essentially, each instance of Alisha becomes tailored to each customer, equipped to handle all inquiries about their company and products,” Shrivastava said.

The startup claims Alisha is the most reliable tool in the market, with ThorV2 near zero hallucination and reliability for function calling at 99.5%, which is impressive. 

The startup also claims other models like Claude-3 Opus and GPT-4o have a reliability score of 59.7% and 83.9%, respectively. 

“When using ChatGPT, if you provide the same prompt twice, chances are you’ll receive different answers each time. This variability is inherent to how language models operate,” Biswas pointed out.

However, this very nature of LLMs and the fact that they hallucinate limit the use of these large models for AI agentic workflows.

“When it comes to tool usage, for instance, accessing your CRM, you want very reliable and deterministic action because you don’t want wrong information to be put in your CRM or send a wrong email to your customer. 

“Mistakes at the LLM level can propagate widely, impacting multiple areas with significant business repercussions. Our model ensures 100% reliability in these critical functions,” Biswas added.

(From left to right: Sarthak Shrivastava and Sudipta Biswas, co-founders at Floworks)

ThorV2 Powers Alisha

Thor2, which powers Alisha, is a mixture of agents (MoA) based on a proprietary LLM architecture built from scratch.

“ThorV2 comprises eight distinct AI agents or LLMs, incorporating both open-source and proprietary models. While some are pre-trained, others require us to rebuild their architecture entirely to align with our specific requirements,” Biswas said.

He further claims that Thorv2 is not only 36% more accurate than OpenAI’s GPT-4o but also 4x cheaper and almost 30% faster in terms of latency.

When asked about the cost of building ThoV2, Biswas revealed that it was around $1 million lower, which is significantly lower compared to foundational models.

Moreover, ThorV2 costs just $1.60 per 1000 queries, making it 175% cheaper than GPT-4o.

(Source: Floworks)

Voice Capabilities in the Pipeline 

The founders also revealed that voice capabilities for their Alisha is something they are working towards. Earlier this year, we saw both OpenAI and Google fascinate everyone with the human-like voice capabilities of their models.

Though OpenAI made GPT-4o available for free, it is yet to release the voice capabilities. Biswas reveals that voice is something they are working towards, but he does not see a demand for multimodal, which involves videos. 

Moreover, he claims Floworks does not want to make bold promises and then take an eternity to deliver a product. 

“It’s very easy to build a prototype and woo the audience, but then to actually build and release a production version, where everyone can use it scalably, that’s a whole different challenge,” Biswas said.

An alumni of IIT Kharagpur, Biswas points out a competitor of Floworks called Adept AI. “The company has raised around $400 million so far, and despite being in existence for over two years, still hasn’t released a product. In fact, the company is on the verge of breaking up,” he added.

Automating End-to-end Sales 

While Alisha is designed for function calling, the vision of the company is to build an AI system that handles end-to-end sales functions in an enterprise.

“We actually envision that in the near future, using Floworks, companies will not require sales teams. It will just be basically creating a good product and then actually training an AI system to sell the product,” Shrivastava said.

When asked if Alisha can make cold calls, he added that Alisha is already sending emails on behalf of humans. Soon, it will start reaching out to people on LinkedIn, WhatsApp and other mediums.

“Voice call too is just another mode of communication, and once the technology is ready, Alisha, too, will start making calls,” Shrivastava added.

While the co-founders are confident, it remains to be seen how much of these can be achieved, particularly whether AI can grasp the nuances of sales functions.

Expansion Plans 

Biswas revealed that the team size is 17 currently and they plan to hire for additional roles and expand the team to around 30-35 in the coming months. 

The company started active sales only two months back and so far has acquired around 14 customers both in the Indian and the US markets. Some of the customers include Anya, Unscript, and Qodex.

“We are growing 100% every month when it comes to customer acquisition, and we are getting a lot of referral customers who are knocking on our doors and enquiring about our product,” Shrivastava said.

Alongside Alisha, Floworks’ upcoming products include AI RevOps, AI project managers, and AI Executive Assistants. These innovations aim to optimise business operations, enabling “one-person unicorns.”

📣 Want to advertise in AIM? Book here

Picture of Pritam Bordoloi

Pritam Bordoloi

I have a keen interest in creative writing and artificial intelligence. As a journalist, I deep dive into the world of technology and analyse how it’s restructuring business models and reshaping society.
Related Posts
19th - 23rd Aug 2024
Generative AI Crash Course for Non-Techies
Upcoming Large format Conference
Sep 25-27, 2024 | 📍 Bangalore, India
Download the easiest way to
stay informed

Subscribe to The Belamy: Our Weekly Newsletter

Biggest AI stories, delivered to your inbox every week.

Flagship Events

Rising 2024 | DE&I in Tech Summit
April 4 and 5, 2024 | 📍 Hilton Convention Center, Manyata Tech Park, Bangalore
Data Engineering Summit 2024
May 30 and 31, 2024 | 📍 Bangalore, India
MachineCon USA 2024
26 July 2024 | 583 Park Avenue, New York
MachineCon GCC Summit 2024
June 28 2024 | 📍Bangalore, India
Cypher USA 2024
Nov 21-22 2024 | 📍Santa Clara Convention Center, California, USA
Cypher India 2024
September 25-27, 2024 | 📍Bangalore, India
discord-icon
AI Forum for India
Our Discord Community for AI Ecosystem, In collaboration with NVIDIA.