Athina AI (YC W23)

We're hiring! We're building an IDE that lets engineering teams develop AI-powered products 10x faster. If you're a highly-driven, talented individual interesting in shaping the future of AI-powered software, then apply to work at Athina. Location: San Francisco

Here's a dynamic database of 50+ AI research papers ?? AI research papers are an excellent resource for staying updated on the latest developments in the AI space. ?? But let’s be honest – we all have countless papers scattered across bookmarks, Excel sheets, PDFs, Notion, and other places in a completely unstructured manner. To solve this, Paras Madan from our team built a dynamic database of these papers, categorized by genre. We’ll be updating it regularly. It includes: ?? Links to all papers ?? Summaries ?? Key highlights And the best part? ?? Since it’s built on AI0 sheets, you can heavily customize it by adding more columns like: ?? LLM prompts ?? API calls ?? Web scrapers & search tools ?? Data extractors ?? Custom code blocks And more... We hope you find this useful! Link in comments ??

Over the past two years, we’ve helped hundreds of teams take their AI applications from?prototype to production—helping them with?testing, monitoring, and iterating?on their AI. Along the way, we’ve noticed something interesting.?One of the most powerful and widely used features of Athina is its AI-powered spreadsheets. Today, we're showcasing some of those examples built using?our no-code interface that we call—AI0 ?? It is an?AI-first spreadsheet?that lets you?automate entire workflows and execute tasks at scale. What can you do with AI0? ? Deep Web Research?– Instantly pull insights from multiple sources ? Data Enrichment?– Automatically add company info, social data & more ? Data Extraction?– Extract key details from documents, web pages, or structured data ? Document Intelligence?– Summarize, categorize, and analyze files effortlessly. What makes AI0 so powerful? ?? ??? Pre-Built AI Tools?– Firecrawl web scraper, Perplexity search, social data extractors, document parsers, Exa & Tavily search, API calls, and more ? ?? Works With Any Dataset?– People, companies, web pages, documents, text—you name it ? ??? Build & Share Custom Blocks?– Anyone can create new AI-powered blocks and publish them for the community Want to see what’s possible? Check out some real-world use cases in the comments ?? If this looks useful, let’s connect—I’d love to hear how AI0 can help your team!?

?? Everyone’s talking about Manus AI , and those with early access are sharing their thoughts. But keeping up with all the discussions on Reddit? That’s a nightmare. So, we built an AI workflow to do the heavy lifting. In just seconds, it analyzes an entire Reddit thread and summarizes the key insights. It can even email the summary to you. ?? How? I simply connected a pre-built Reddit analyzer tool with an LLM tool, added a quick prompt on Athina AI (YC W23)—and boom, done! If you’re building AI workflows, we’ve got a library of pre-built tools to make it effortless. ?? Check out the workflow in the comments!

Models are getting better at reasoning every day. We’ve seen our clients throw massive documents at these models—extracting insights, detecting risks, and uncovering patterns in seconds. ?? When it comes to reasoning tasks, accuracy >> latency. One of the most powerful use cases? Reviewing large contracts. ?? AI can critique documents, suggest changes, and highlight risks in minutes—what used to take hours. Here’s a simple 2-step AI workflow you can use to: ? Extract key clauses ? Detect potential risks ? Highlight critical terms ? Ensure compliance with legal standards If you're working in a regulated industry or care about privacy, you can even perform document intelligence using open source models hosted on your VPC. ?? Link in comments. Try it out! ??

This week's research breakthroughs pushed the boundaries of RAG, AI Agents, and LLM Evaluations ???? We’ve compiled a list of the Top 10 LLM Papers from the past week shaping these areas: 1?? Knowledge Graph-Guided Retrieval Augmented Generation 2?? Fairness in Multi-Agent AI: A Unified Framework for Ethical and Equitable Autonomous Systems 3?? Preventing Rogue Agents Improves Multi-Agent Collaboration 4?? CODESIM: Multi-Agent Code Generation and Problem Solving through Simulation-Driven Planning and Debugging 5?? Forget What You Know about LLMs Evaluations - LLMs are Like a Chameleon 6?? BenchMAX: A Comprehensive Multilingual Evaluation Suite for Large Language Models 7?? Single-Agent Planning in a Multi-Agent System: A Unified Framework for Type-Based Planners 8?? Commercial LLM Agents Are Already Vulnerable to Simple Yet Dangerous Attacks 9?? Ask in Any Modality: A Comprehensive Survey on Multimodal Retrieval-Augmented Generation (RAG) ?? ParetoRAG: Leveraging Sentence-Context Attention for Robust and Efficient Retrieval-Augmented Generation Ready to dive deeper into how these papers are driving the next wave of AI innovation? ?? Read the full blog in the first comment ??

This week’s Research highlights and brings exciting developments in RAG, AI Agents, and LLM Evaluations. We’ve compiled a list of the Top 10 LLM Papers from the past week shaping these areas: 1?? The AI Agent Index: Public Database for AI agent architectures 2?? Learning to Plan & Reason for Evaluation with Thinking-LLM-as-a-Judge 3?? Training an LLM-as-a-Judge Model: Pipeline, Insights, and Practical Lessons 4?? GFM-RAG: Graph Foundation Model for Retrieval Augmented Generation 5?? Multi-Agent Design: Optimizing Agents with Better Prompts and Topologies 6?? Rethinking Mixture-of-Agents: Is Mixing Different Large Language Models Beneficial? 7?? Enhancing Online Learning Efficiency Through Heterogeneous Resource Integration with a Multi-Agent RAG System 8?? ScoreFlow: Mastering LLM Agent Workflows via Score-based Preference Optimization 9?? DeepRAG: Thinking to Retrieval Step by Step for Large Language Models ?? Agentic Reasoning: Reasoning LLMs with Tools for the Deep Research Ready to dive deeper into how these papers are shaping the landscape of LLM research and applications? ?? Read the full blog in the first comment ??

Agentic RAG using DeepSeek AI - Qdrant - LangChain ?? [Open-source notebook] If you're looking to implement Agentic RAG using DeepSeek's R1 model we've published a ready-to-use Colab notebook (link in comments). ?? ???This notebook uses an agentic Router along with RAG to improve the retrieval process with decision-making capabilities. ??? It has 2 main components: 1?? Agentic Retrieval The agent (Router) uses multiple tools—like vector search or web search—and decides which to invoke based on the context. 2?? Dynamic Routing It maps the optimal path for retrieval— Retrieves data from vector DB for private knowledge queries and uses web search for general queries! ?? Whether you're building enterprise-grade solutions or experimenting with AI workflows, Agentic RAG can improve your retrieval processes and results. ?? What advanced technique should we cover next?

How well does DeepSeek AI R1 work on your data? ?? Everyone is talking about Deepseek — breaking benchmarks and outpacing competitors, even the proprietary ones. And truly, it is Fire! ?? But here’s the thing: benchmarks are one thing, and your use case is another. If you’re dealing with complex use cases like: ?? Legal document analysis ?? Financial risk assessment ?? Scientific research summarization ?? Coding assistance you need a powerful model with top-tier reasoning capabilities. For simpler tasks like routine data extraction, keyword-based document retrieval, or basic summarization, other models might get the job done just fine. The best way to find out? Try it for yourself. We rolled out Deepseek reasoning support over the weekend ?? Give it a shot — Athina AI (YC W23) + Deepseek https://app.athina.ai

?? While building a large 24-step AI pipeline, we wanted to find the best model b/w GPT 4o, Claude Sonnet, Gemini 2.0, and Deepseek R1 for the reasoning task. It literally took us a few minutes to build the entire pipeline and compare all the models side by side.??? Typically, with such experiments, you need to take care of a lot of dimensions - type of task, model performance, model size, LLM metrics (cost, latency), inference efficiency, etc.??? It could be a nightmare to do all of this on spreadsheets! This is why you need a unified platform that can help you—from building multi-step pipelines to testing models side-by-side and finding the perfect setup for your use case. If you’re struggling with the same challenges, let’s chat! I’d love to help your team by building a custom AI pipeline you can test with your own datasets, models, and configurations.