Athina is a collaborative AI development platform designed for teams to build, test, and monitor AI features.

Athina

Athina AI (YC W23) | 3,288 followers on LinkedIn. A data-centric IDE for teams to prototype, experiment, evaluate and monitor production-grade AI | Athina helps LLM developers prototype, experiment, evaluate and monitor production-grade AI pipelines.

Co-founder Athina AI (Y Combinator W23)

Himanshu Bamoria

Building an IDE for GenAI development | Co-founder, Athina AI (YC W23) | Ex-Google, YC

Shiv Sakhuja

Founding Engineer@Athina.ai

Akhil Bisht

AI Developer Advocate @ Athina AI

Prasad Mahamulkar

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

We are thrilled to be featured in The American Bazaar, sharing the news of our recent $3M funding round! A big thank you to The American Bazaar for highlighting our journey as we continue to build Athina AI, an IDE crafted for AI development teams. Check out the full story here: https://lnkd.in/gXhFSCAx Grateful for the support and excited about what lies ahead!

A huge shout-out to YourStory Media and Bhuvana Kamath for featuring our recent fundraise! We're thrilled to be recognized and excited about the journey ahead. This is a big moment for us at Athina AI, and we're grateful to everyone who's supported us so far. More to come as we continue building tools for AI development teams. Check out the story here: https://lnkd.in/gNDP_jV4

🚀 We're thrilled to share that Athina AI has successfully raised a $3M seed round! 💡 This funding fuels our mission to accelerate the development of an end-to-end AI platform that empowers teams to manage prompts, models, and datasets in production—addressing some of the toughest challenges in AI engineering today. 🤝 We're proud to be partnering with companies like Perplexity and Doximity , helping them build production-grade AI applications that make a real difference. 🎉 Backed by Y Combinator , Kleiner Perkins , Scout Fund, GSF , Bharat Founders Fund , and incredible strategic angel investors, we're more committed than ever to delivering impactful solutions. 🌟 The groundwork is laid, and we're excited to keep building—thanks for being part of this journey with us!

Thrilled to announce that Athina AI (YC W23) has raised a $3M seed round! 🚀 Here’s what makes our platform stand out: • We’re tackling a crucial pain point in AI development – simplifying the complexity of experimenting with prompts, models, and datasets at production scale • We’re already supporting incredible customers like Perplexity, Doximity, and Meesho in building reliable AI systems • Backed by YC, Kleiner Perkins Scout Fund, Snorkel AI's CEO, and others who believe in our vision Tools that streamline the AI development lifecycle are becoming more and more essential. We’re excited to be positioned as a key player in this rapidly advancing field. Kudos to the Athina AI team and big thanks to our supporters for helping us reach this milestone. Onward! 💪

🚀 Excited to announce Athina AI (YC W23) 's fresh $3M seed fundraise! 🚀 We’re incredibly grateful for the support of amazing backers like Y Combinator , Flourish Ventures , Kleiner Perkins Scout Fund, GSF , Bharat Founders Fund , and notable angels including Alexander Ratner (CEO, Snorkel AI) and Denis Yarats (CTO, Perplexity). This is a major milestone toward our mission of enabling companies to build production-grade AI applications! AI engineering requires extensive experimentation with prompts, models, datasets, and handling complex, unpredictable responses. Athina’s end-to-end AI development platform is designed to help teams overcome these challenges and succeed. We’re proud to support AI teams from unicorns like Perplexity, Doximity, Meesho, and PhysicsWallah in building reliable, production-ready systems. The past few months have brought phenomenal growth, with a surge in new users and even more successful customer outcomes. None of this would be possible without our incredible team at Athina. Huge shoutout to Akhil Bisht , Vivek Aditya , Omid Elliyoun, Cody, Prasad Mahamulkar , Paras Madan , John Carlo Sola , Aamir Nabi Badyari and my exceptional co-founder Shiv Sakhuja for their relentless dedication! 🙌 A heartfelt thanks to our investors, advisors, and everyone who has been a strong support throughout this journey. This is just the beginning! 🔥

Super excited to share that Athina AI (YC W23) has closed a $3M seed round! This is a major milestone in our mission to enable teams to build production-ready AI applications and agents. AI will be used in every software product in the next 5 years—but building production-ready AI is very hard. Athina AI (YC W23) helps teams collaborate to build, test and monitor AI applications for production. The momentum has been incredible. In just a few months, we’ve grown rapidly and are proud to support world-class AI teams, including unicorns like Perplexity, Doximity, Meesho, and PhysicsWallah. We're also thrilled to be backed by some amazing investors including Y Combinator , Flourish Ventures , Kleiner Perkins Scout Fund, GSF , Bharat Founders Fund , Alexander Ratner (CEO, Snorkel AI) and Denis Yarats (CTO, Perplexity), among others. Incredibly grateful to our team at Athina ❤️ The future of software is AI… and we’re just getting started. P.S - we're hiring! (link in comments)

💡 If you're working with large, complex documents, the Parent Document Retriever (PDR) technique could be useful. It is generally used for maintaining both specific details and the overall context of the documents. 🔹 PDR splits large documents into smaller pieces, called "child chunks". These chunks are stored in a way that lets the system find and compare specific parts of a document with a user’s query. The large document, or "Parent," is still kept but is only retrieved if one of its child chunks is relevant to the query. 📖 Prasad Mahamulkar has written an in-depth guide about Advanced RAG, covering how PDR can improve document retrieval. Why is this helpful? 1️⃣ Better Embedding Similarity Large documents often cover several topics, so using one vector to represent them isn’t ideal. By breaking the document into chunks, each chunk can be matched more precisely to the query. 2️⃣ Better Context Retrieval The small child chunk alone doesn’t give enough information. Returning the full parent document provides the surrounding context that can make the response more comprehensive and meaningful. Comparing the Context Precision of PDR vs your existing RAG could be a good indicator to check if this is beneficial for your use case. Here's how to do it 👇

Zillow uses LLMs to power a guard railing system that prohibits discriminatory/offensive user queries and content 🚀 Their system is built on 3 key components each having its pros and cons: 1️⃣ Prompt Engineering Zillow’s prompts guide the AI to be fair and avoid discrimination, especially on sensitive topics. However, the system became overly cautious, often declining safe questions, which hurt user experience by withholding helpful information. In other words, the system has high recall but sub-optimal precision. 2️⃣ Stop List: To complement the prompts, a database of words linked to protected classes was created. This stop list acted as a lexical check, flagging queries with non-compliant words. Although effective, it fell short on handling nuanced, context-dependent phrases. 3️⃣ Classifier Model (BERT): For flexibility and speed, Zillow employed a classifier model trained on labeled data specific to fair housing. They used data augmentation techniques to improve model accuracy, ensuring it could accurately detect sensitive queries while enhancing real-time processing. ✅ The Combined Approach Zillow brought it all together with a guardrail system that integrates prompt engineering, stop lists, and fast-inference classifiers. This created a robust API to validate both user input and system responses 🌐 Link to complete case study in the comments 👇

✨ Answer Semantic Similarity refers to the similarity between the embedding of the RAG response and the embedding of the Ground truth answer. Measuring the semantic similarity can offer valuable insights into the quality of the generated response. The score ranges from 0 to 1, with higher scores showing a better match between the two answers. Here's how to calculate the semantic similarity score using ragas 👇

ATHI

AI:ATHINA

Athina