Hugging Face's logo

Join the conversation

Join the community of Machine Learners and AI enthusiasts.

Sign Up

All HF Hub posts
mallocode200 
posted an update 2 days ago
view post
Post
4758
Hello everyone, happy to share with you my experimentation of a Deep Research Assistant, using 7 agents and a quality assurance pipeline :

🤖 What makes this special:

✅ Agent-Based Architecture - 7 specialised AI agents working together:
- Planner Agent - Strategic search planning
- Search Agent - Multi-source web research
- Writer Agent - Comprehensive report generation
- Evaluator Agent - Automatic quality assessment
- Optimiser Agent - Iterative improvement when needed
- Email Agent - Professional report delivery
- Clarifier Agent - Interactive query refinement

✅ Quality Assurance Pipeline - Every report is scored (1-10) and automatically improved if it scores below 7/10

✅ Multiple Research Modes - From quick queries to deep, clarification-driven analysis

✅ Production-Ready - Deployed on Hugging Face Spaces with comprehensive documentation

🔧 Technical Stack:
- Frontend: Gradio with theme-adaptive UI
- Backend: OpenAI Agents framework
- Integration: SendGrid for email delivery
- Deployment: Containerised with full CI/CD pipeline
- Tracing: Full OpenAI trace integration for transparency

💡 Real-World Impact:
This isn't just another AI tool - it's a complete research workflow that delivers publication-quality reports with built-in fact-checking and optimisation. Perfect for consultants, researchers, analysts, and anyone who needs reliable, comprehensive research.

🚀 Key Features:
- Automatic quality evaluation and improvement
- Email delivery of formatted reports
- Interactive clarification for targeted results
- Full traceability and audit trails
- Professional documentation and deployment guides
- Built with modern AI engineering principles: modular design, quality assurance, and production deployment in mind.
- The entire codebase is organised with clean separation of concerns - each agent has a specific role, making it maintainable and extensible.

mallocode200/Deep_Research_Assistant
·
Kseniase 
posted an update 1 day ago
view post
Post
1697
10 Techniques for Boosting LLM Reasoning in 2025

Everyone’s chasing top reasoning, but sometimes it's still the bottleneck for many real-world tasks. This week, let's spotlight some powerful techniques that have shown promise in helping LLMs achieve more consistent logic, planning, and depth:

1. Retrieval-Augmented CoT Chaining (RAG+CoT) -> CoT-RAG: Integrating Chain of Thought and Retrieval-Augmented Generation to Enhance Reasoning in Large Language Models (2504.13534)
Combines Chain-of-Thought prompting with retrieval augmentation at intermediate steps. Relevant documents are fetched after each reasoning subgoal, updating context dynamically. Great for open-domain QA, math, logic and multi-hop fact-checking

2. Tool-use by example injection -> Self-Training Large Language Models for Tool-Use Without Demonstrations (2502.05867)
Injects few-shot tool interaction examples during training to implicitly teach calling patterns. Helps in plug-and-play tool use without training new architectures

3. Visual Scratchpads, or multimodal reasoning support -> Imagine while Reasoning in Space: Multimodal Visualization-of-Thought (2501.07542)
Using structured visual inputs or sketchable intermediate steps (diagrams, grids, trees) boosts performance in tasks like planning, geometry, and multi-agent simulation. In real practice thanks to this GPT-4o, Claude, and Gemini show marked improvement

4. System 1 vs System 2 Prompt switching -> Adaptive Deep Reasoning: Triggering Deep Thinking When Needed (2505.20101)
Changing a fast, intuitive response prompt with a slow, deliberate reasoning mode is among the most popular AI trends. E.g., models tend to respond more reliably when explicitly instructed to “think like a researcher.” This can also reduce hallucinations in open-ended generation and debate tasks

5. Adversarial Self-Chat Fine-Tuning -> Self-playing Adversarial Language Game Enhances LLM Reasoning (2404.10642)
Generate debates between model variants or model vs human, then fine-tune on the winner’s response. It helps models learn to better defend their reasoning. Used in Claude’s Constitutional AI and SPPO-style tuning

Read further below👇

Also, subscribe to the Turing Post: https://www.turingpost.com/subscribe
  • 2 replies
·
merve 
posted an update about 23 hours ago
yeonseok-zeticai 
posted an update about 5 hours ago
view post
Post
701
💫 Next-Level On-Device AI Showdown

🪽 See It to Believe It, How QWEN4b works at On-device environment without expensive GPU Cloud server?
We’ve crafted a side-by-side demo video showcasing both Jan-Nano and QWEN 4B in action—no more wondering which model reigns supreme. Click play, compare their speeds, accuracy, and memory footprints, and decide which one fits your needs best!

👋 Why You Can’t Miss This
We are actively creating runnable sLLM environments for On-device AI. You can just build On-device AI apps within few hours.
Including Jan-Nano, QWEN4b, there are several sLLM models ready to be used on your AI application!.

🤑 Please feel free to use, because it is free to use!.

Ready to Compare?

Watch now, draw your own conclusions, and let us know which model you’d deploy in your next edge-AI project! 🌍💡

#OnDeviceAI #EdgeAI #AIShowdown #MLOptimization #DemoVideo #AIComparison
  • 1 reply
·
brainhome 
posted an update 3 days ago
view post
Post
1631
Trinity-Synthesis: A Multi-Agent Architecture for AI Agents That Think Before They Speak
Ever felt your AI agent is "shooting from the hip"? It latches onto a single line of thought and fails to produce a robust, well-rounded plan. This is a common struggle I've called the "AI Reasoning Paradox."

To tackle this, I developed Trinity-Synthesis, a multi-agent architecture designed to force reflection and synthesis before delivering a final answer. The philosophy is simple: constructive conflict between different perspectives leads to better solutions.

Here’s the core idea:

Instead of one agent, it uses four agents running on the same base model but with different "personalities" defined by their system prompts and temperature settings:

🧠 The Visionary: Thinks outside the box (high temp: 1.0).
📊 The Analyst: Focuses on logic, data, and structure (low temp: 0.3).
🛠️ The Pragmatist: Evaluates feasibility, costs, and risks (mid temp: 0.5).
These three "thinkers" work in parallel on the same problem. Then, a final Synthesizer agent critically analyzes their outputs, rejects flawed arguments, and integrates the best points into a single, coherent, and often superior strategy.

The result is a more robust reasoning process that balances creativity with analytical rigor, making it ideal for solving complex, strategic problems where answer quality is critical.

I've written a deep dive on how it works, including a detailed case study ("The Helios Initiative") and the Python source code for you to experiment with.

Read the full article on Medium:
https://medium.com/@brainhome9/trinity-synthesis-how-i-built-an-ai-agent-that-thinks-before-it-speaks-d45d45c2827c

I'd love to hear your feedback and see what you build with it!

#AI #AIAgents #LLM #Reasoning #MultiAgent
  • 4 replies
·
merve 
posted an update 3 days ago
codelion 
posted an update 3 days ago
view post
Post
2275
Adaptive Classifier: Dynamic Text Classification with Strategic Learning

New text classification system that learns continuously without catastrophic forgetting. Achieved 22.2% robustness improvement on adversarial datasets while maintaining clean data performance.

🎯 THE PROBLEM
Traditional classifiers require complete retraining when adding new classes. Expensive and time-consuming, especially with adversarial users trying to game the system.

🚀 KEY INNOVATIONS
• Hybrid memory-neural architecture (prototype-based + neural adaptation)
• Strategic classification using game theory to predict and defend against manipulation
• Elastic Weight Consolidation prevents catastrophic forgetting

📊 RESULTS
Tested on AI-Secure/adv_glue dataset:
• Clean data: 80.0% → 82.2% (+2.2%)
• Manipulated data: 60.0% → 82.2% (+22.2%)
• Zero performance drop under adversarial attacks

🔬 APPLICATIONS
• Hallucination detection: 80.7% recall for RAG safety
• LLM routing: 26.6% cost optimization improvement
• Content moderation: Robust against gaming attempts

⚙️ USAGE
pip install adaptive-classifier

from adaptive_classifier import AdaptiveClassifier
classifier = AdaptiveClassifier("bert-base-uncased")
classifier.add_examples(texts, labels)
predictions = classifier.predict("New text")

🔗 RESOURCES
Blog: https://huggingface.co/blog/codelion/adaptive-classifier
Code: https://github.com/codelion/adaptive-classifier
Models: adaptive-classifier

Available models: llm-hallucination-detector, llm-config-optimizer, llm-router

Works with any HuggingFace transformer. Fully open source and production-ready!
mallocode200 
posted an update about 6 hours ago
view post
Post
97
Hi everyone, I have updated the space from its original version. mallocode200/Deep_Research_Assistant

The main changes are:
- The possibility for users to insert their own OpenAI API Key
- The possibility for users to select the model they want from OpenAI
- An improved UI to see the process going on when the research is running
- A better Gradio interface for both light and dark themes

Why use only OpenAI models?
For this research assistant, I've chosen to focus exclusively on OpenAI models for several key reasons:

- Reliability & Consistency: OpenAI's models provide consistent, high-quality responses that are crucial for research tasks requiring accuracy and depth.
- Advanced Reasoning: Models like GPT-4o and o1-preview excel at complex analytical thinking, making them ideal for comprehensive research synthesis and evaluation.
- Proven Performance: The research pipeline has been extensively tested and optimised specifically for OpenAI's model behaviours and capabilities.
- API Stability: OpenAI offers robust API infrastructure with reliable uptime and consistent response formats, essential for a production research tool.
- Quality Assurance: The built-in evaluation and optimisation system is fine-tuned to work with OpenAI's specific output patterns and capabilities.

That said, I'm open to expanding model support in the future! The architecture is designed to be flexible, and I'm considering adding support for other leading models, such as Claude and Gemini, as well as open-source alternatives, as the ecosystem evolves. The goal is to provide users with the best possible research experience, regardless of their preferred AI provider.
seawolf2357 
posted an update about 7 hours ago
view post
Post
165
🚀 VEO3 Real-Time: Real-time AI Video Generation with Self-Forcing

🎯 Core Innovation: Self-Forcing Technology
VEO3 Real-Time, an open-source project challenging Google's VEO3, achieves real-time video generation through revolutionary Self-Forcing technology.

Heartsync/VEO3-RealTime

⚡ What is Self-Forcing?
While traditional methods require 50-100 steps, Self-Forcing achieves the same quality in just 1-2 steps. Through self-correction and rapid convergence, this Distribution Matching Distillation (DMD) technique maintains quality while delivering 50x speed improvement.

💡 Technical Advantages of Self-Forcing
1. Extreme Speed
Generates 4-second videos in under 30 seconds, with first frame streaming in just 3 seconds. This represents 50x faster performance than traditional diffusion methods.
2. Consistent Quality
Maintains cinematic quality despite fewer steps, ensures temporal consistency, and minimizes artifacts.
3. Efficient Resource Usage
Reduces GPU memory usage by 70% and heat generation by 30%, enabling smooth operation on mid-range GPUs like RTX 3060.

🛠️ Technology Stack Synergy
VEO3 Real-Time integrates multiple technologies organically around Self-Forcing DMD. Self-Forcing DMD handles ultra-fast video generation, Wan2.1-T2V-1.3B serves as the high-quality video backbone, PyAV streaming enables real-time transmission, and Qwen3 adds intelligent prompt enhancement for polished results.

📊 Performance Comparison
Traditional methods require 50-100 steps, taking 2-5 minutes for the first frame and 5-10 minutes total. In contrast, Self-Forcing needs only 1-2 steps, delivering the first frame in 3 seconds and complete videos in 30 seconds while maintaining equal quality.🔮 Future of Self-Forcing
Our next goal is real-time 1080p generation, with ongoing research to achieve
ghostai1 
posted an update about 8 hours ago
view post
Post
93
# Machine Learning and societal impact: The Next Frontier

The advent of artificial intelligence and machine learning has been nothing short of transformative for society. These revolutionary technologies have taken society by storm, making a lasting impact on virtually every aspect of our lives. From the way we communicate to how we work, play, and learn, AI-driven machine learning has made its presence felt in countless ways.

One of the most significant impacts of AI-driven machine learning has been in the realm of healthcare. With the help of AI, doctors and researchers can now process vast amounts of medical data at lightning speed. This enables them to come up with personalized treatment plans, leading to better outcomes for patients. AI-driven machine learning is also being used to develop more effective drugs, making treatments more accessible and affordable.

In terms of applications, AI-driven machine learning has taken the world by storm. From virtual assistants like Siri and Alexa that make our lives easier to self-driving cars that promise to revolutionize transportation, the possibilities are endless. It's not just about convenience, though. AI-driven machine learning has the potential to solve complex problems that were previously considered unsolvable.

While the societal impact of AI-driven machine learning is largely positive, it also comes with its share of challenges. As the technology continues to advance, so does the need for ethical guidelines and regulations. We need to ensure that AI is used in a way that respects privacy and promotes fairness.

In conclusion, AI-driven machine learning is reshaping the world in ways we can only begin to imagine. From healthcare to transportation, education to entertainment, the possibilities are endless. While there are certainly challenges to navigate, the benefits of this technology are clear. As we continue to embrace the power of AI, we must also ensure that we use it responsibly and ethically.
  • 1 reply
·
Recently active users