Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Hugging face x mangoes.ai case study #1572

Closed
wants to merge 3 commits into from
Closed
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
11 changes: 10 additions & 1 deletion _blog.yml
Original file line number Diff line number Diff line change
Expand Up @@ -2968,4 +2968,13 @@
- open-source-collab
- onnxruntime
- onnx
- inference
- inference

- local: mangoes.ai-case-study
title: "Transforming Behavioral Health Equity with Audio ML and LLMs: Mangoes.ai and Hugging Face Fast-Track Prompt Engineering, Supervised Fine-Tuning and RLHF"
author: on1onmangoes
guest: true
thumbnail: /blog/assets/78_ml_director_insights/mangoes.png
date: October 13, 2023
tags:
- case-studies
Binary file added assets/78_ml_director_insights/mangoes.png
Loading
Sorry, something went wrong. Reload?
Sorry, we cannot display this file.
Sorry, this file is invalid so it cannot be displayed.
88 changes: 88 additions & 0 deletions mangoes.ai-case-sudy.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,88 @@
---
title: "Transforming Behavioral Health Equity with Audio ML and LLMs: Mangoes.ai and Hugging Face Fast-Track Prompt Engineering, Supervised Fine-Tuning and RLHF"
thumbnail: /blog/assets/78_ml_director_insights/mangoes.png
authors:
- user: on1onmangoes
---

<!-- {blog_metadata} -->
<!-- {authors} -->

# Transforming Behavioral Health Equity with Audio ML and LLMs: Mangoes.ai and Hugging Face Fast-Track Prompt Engineering, Supervised Fine-Tuning and RLHF

## Introduction

The ability to scrutinize nuances in acoustics, voice, and speech allows mangoes.ai to identify early signs of mental health concerns, thereby helping healthcare providers take proactive steps in treatment planning. This capability is especially crucial in serving underserved heroes in the community who face mental wellness challenges. Through meticulous data analysis and real-time alerts, we not only save lives, we can prevent institutional biases and ensure that mental health treatment is more equitable and effective for all.

Our collaboration with Hugging Face has turbocharged these capabilities, helping us reach an unprecedented level of accuracy and reliability in symptom detection and risk monitoring.

#### _"Working with such visionary yet approachable professionals has been a humbling joy, aiding in our company's growth trajectory. Special appreciation goes to the Hugging Face Team and our supportive colleagues who are truly magnificent in areas such as Audio and Generative AI"_
-- Amit Lamba, Mangoes.ai Founder & CEO.

<p align="center">
<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/blog/llama2-non-engineers/mangoes1.png"><br>
</p>


## The Challenge: Scalability Without Sacrificing Quality

As demand grew for real-time mental health solutions, Mangoes.ai faced hurdles in:

- Cerebral Ideation to harness Generative and Artificial Intelligence: Why, When and How Commercialization Innovation in Machine Learning Models
- Scaling products without compromising accuracy
- Achieving speed for instant insights from complex data
- Eliminating biases in mental health diagnoses
- Enhance Retrieval Augmented Generation to improve veracity using PineCone, Selenium and cutting edge pipelines

Mangoes.ai is on the forefront of therapeutic AI, particularly in the field of mental health. With clients demanding increasingly sophisticated, real-time solutions, the challenges of scale, accuracy, and speed loomed large.

To navigate these complexities, the team at Mangoes.ai sought expert guidance, recognizing that an infusion of specialized knowledge would significantly elevate their product offerings. However, the company was grappling with several obstacles—most notably, the intricacies of scaling without compromising on accuracy, privacy, and accessibility. They needed to evolve quickly in an ever changing tech landscape. Moreover, they aimed to eliminate institutional biases in mental health diagnoses, a feat that required an incredibly nuanced understanding of human language and sentiment.


## The Hugging Face Solution: Expertise Meets Execution

Upon embarking on this journey, Mangoes.ai gained unprecedented access to the intellectual capital of Hugging Face's team, which included eminent professionals like Yifeng Yin and Sanchit Gandhi as well as Gradio’s founding engineering team. These experts became integral contributors to Mangoes.ai's strategic vision, offering hands-on guidance that streamlined processes across the board:

- Accelerating Acoustic Model Development: Rather than facing the arduous task of building an acoustic model from the ground up, Yifeng Yin proposed the use of pre-trained models differentiated with highly sophisticated and cutting-edge Supervised Fine Tuning and complemented with LLMs with Reinforced Learning using Human and AI Feedback, effectively cutting months off the development cycle. The correct solution for the problem saved us time and also GPU related costs and resources.
- Revolutionizing Speaker Diarization: Sanchit Gandhi's insights on the synergy between Whisper JAX and Pyannote audio provided a paradigm shift in Mangoes.ai's approach to speaker diarization, leading to groundbreaking results.
- Prompt Engineering & Performance: Under the aegis of the Hugging Face team, Mangoes.ai explored the transformative power of Generative AI and prompt engineering techniques. This considerably lifted the performance metrics of their deployed solutions.
- Data Heterogeneity, Interoperability and Actionability: Built proprietary Generative capabilities to allow data to be acted upon by clinicians and patients. Each hospital, and individual therapist caters to different clienteles and also use electronic medical data that is not consistent
- Power to Deliver: Beyond the intellectual resources, Hugging Face provided Mangoes.ai with high-powered A100 GPUs, offering a drastic reduction in the time required for model training and inference.


## The Results: A Success Story Unfolds

Whether from Hugging Face’s sales, client or Machine Learning + Gradio team, the approachability, commitment to mangoes.ai success brings about exciting, objective and fascinating discussions. The collaboration quickly led us to design and deploy value generating and commercially viable product suite to empower our clients and users.Their network is not only deep but also highly effective.

By integrating AI-powered psychiatric evaluations, self-awareness exercises, multimodal conversational insights, and blazing fast transcriptions, mangoes.ai’s collaboration enhances user adoption and fosters improved outcomes for communities in need. With this comprehensive suite of tools, mangoes.ai sets new standards in mental health care, exemplifying their commitment to the well-being of their diverse user base.

Each of our product is able to deliver SOTA or best in class performance with Hugging Face’s guidance. At Mangoes.ai, our roadmap is clear and ambitious. Our primary goal is to create a powerful yet ethical AI framework that addresses specific use-cases in mental health symptomatology and therapy. The first step involves identifying the most effective models and techniques that align with these objectives. From there, we transition into building robust Deep Learning applications, emphasizing best practices every step of the way. Alongside these technical endeavors, we also delve into specialized tasks like Semantic Search, Few-Shot Learning, and Model Optimization. But we're not just about algorithms; we're committed to addressing the ethical dimensions and biases within AI. Through these focused efforts, we aim to lead in therapeutic AI, offering not just technology but a responsible approach to global well-being. All of this was done in an accelerated and systematic manner thanks to the Hugging Face team.

<p align="center">
<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/blog/llama2-non-engineers/mangoes2.png"><br>
</p>


The Expert Acceleration Program wasn't merely an add-on service; it became a cornerstone in Mangoes.ai's machine learning strategy. Here's how:

- Speed and Efficiency: With the integration of specialized models, Mangoes.ai managed to reduce their speech recognition time by an astounding 70%.
- Precision at Scale: The innovative speaker diarization process now boasts a whopping 90% accuracy, which is nothing short of industry-leading.
- Quality of Summaries: The implementation of prompt engineering led to a remarkable 35% improvement in the relevance of language model-generated summaries, as corroborated by ROUGE scores.
- VizOps: Users can provide natural language prompts to control the visual style, tailoring charts to be more memorable, engaging, and aesthetic. This allows mangoes.ai to present insights in a format that resonates with diverse audiences. Clinicians can dynamically adapt visualizations to suit their own preferences, as well as individual patients' needs and backgrounds. Rather than a one-size-fits-all approach, infographics can be personalized for maximum comprehension and impact.
- Compute Power: Access to A100 GPUs meant model training times for Audio ML were cut by fourfold, while inference latency was halved. This also allowed for a seamless user experience for demo’s POCs and MVPs

<p align="center">
<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/blog/llama2-non-engineers/mangoes3.png"><br>
</p>


## The Bigger Picture: Why Every Enterprise Should Consider Hugging Face's Expert Acceleration Program

#### _“Investing in the development of psychiatric specialized Large Language Models (LLMs) is undoubtedly the strategic way forward, especially when compared to relying solely on closed LLMs accessed via an API. While it's possible to rapidly prototype using off-the-shelf models like Claude or GPT-4, the advantage of fine-tuning LLMs on high-quality, domain-specific data can't be overstated”_
-Amit Lamba, Mangoes.ai Founder & CEO.

For companies at every juncture—whether fledgling startups or established giants—the program offers a golden opportunity. From fine-tuning models and optimizing latency to deploying solutions on cloud platforms, the Expert Acceleration Program is a holistic solution for businesses that aim to be at the forefront of machine learning advancements.

In an industry where change is the only constant, Hugging Face's Expert Acceleration Program offers not just a roadmap, but a veritable compass, guiding enterprises like Mangoes.ai through the labyrinthine journey that is machine learning. It's not merely about surviving in a competitive environment; it's about thriving, innovating, and setting new industry standards to positively impact millions of lives.

_If you’re feeling inspired, but still need technical support to get started, feel free to reach out and apply for support [here](https://huggingface.co/support#form). Hugging Face offers a paid Expert Advice service that might be able to help._