Vasu Jain

If you are attending ICPR 2024 in the next 5 days (Dec 01-05), then do visit our poster titled: "waLLMartCache: A Distributed, Multi-Tenant and Enhanced Semantic Caching System for LLMs”; my co-author Binay Gupta will be presenting it. Our system waLLMartCache advances the GPTCache framework by incorporating the following features: 1. We introduce the support for a new database Redis in GPTCache – this is used as L2 cache in our designed system. 2. Currently, GPTCache is implemented to be run on a single node which we enhance to span across multiple nodes to handle industry-scale requests and consequently, we also designed a distributed eviction manager. 3. We further create partitions for individual tenants (clients) so that these can be hosted together while maintaining semantic separations. 4. We present a decision engine (i.e., an enhanced semantic similarity checker) that decides whether to cache an LLM response based on retail business use-cases. 5. We showcase that loading FAQs (which can be set to be stored persistently in the memory) while booting the LLM cache can be a simple yet effective strategy to boost cache hits significantly. Some of these contributions have already been merged with the open-source GPTCache code. A big shout out to my other co-authors: Soumik Dasgupta, Anurag Wagh, Lalitdutt Parsai, Geet Vudata, Shally Sangal, Sohom Majumdar, Hema Rajesh, Anirban Chatterjee. #GenAI #LLMs #MachineLearning #ArtificialIntelligence #Industry #Research #Conference #Paper

116

3 Comments

Generative AI has been the talk of the town in recent years. Its paramount ability to efficiently solve diverse business problems using minimal resources and time has attracted the attention of leading market players. We've witnessed a flood of AI products being launched recently. ChatGPT has become my favorite tool of late. It can save you hours if utilized correctly. It can be employed for product recommendations, content creation, and myriad other tasks. By customizing the fundamental model, we can develop various generative AI assistants to address these use cases. I invite you to read my new blog on developing a custom generative AI assistant using AWS services 📰 #ai #generativeai #aws https://lnkd.in/eXAdYdRC

15

Another week ends with a new project Here's 𝗦𝗶𝗹𝗸𝗯𝗼𝘁, A RAG based customer care chatbot for fashion brands. It's the our third project at Headstarter SWE Fellowship. Kudos to my team member Subhan Qamar 👈 for his efforts and dedication. 𝗟𝗶𝘃𝗲 𝗟𝗶𝗻𝗸 🔗 : https://lnkd.in/drXsZJDW 𝗚𝗶𝘁𝗵𝘂𝗯 𝗥𝗲𝗽𝗼🔗 : https://lnkd.in/db_YpzVy 𝗕𝗲𝗵𝗶𝗻𝗱 𝘁𝗵𝗲 𝗦𝗰𝗲𝗻𝗲𝘀: 🏵 1. We've convert dummy data of fashion store into embeddings using ChatGPT embeddings and store in pinecone. 2. When user query something, it's query convert to embeddings and we fetch relevant data from pinecone. 3. We feed the prompt and fetched data to Llama and it gives output. 𝗧𝗲𝗰𝗵 𝗦𝘁𝗮𝗰𝗸: 🙌 - Next.js - Tailwind CSS - Pinecone - Llama API - ChatGPT Embeddings Model - AIMLApi 👊 Provide us feedback, there can be issues in design because we didn't have designer yet in our team. Check it out before Llama API limit exceed... GO GO GO! 👏 Thanks to Bill Zhang for amazing tutorials and Yasin Ehsan 🚀 for providing such a amazing platform.👏 #headstarter #chatbot #llama #embeddings #pinecone #next.js

48

37 Comments

🎉 Great News! 🎉 I'm thrilled to share that our(Animesh Mukherjee, Paramita Das and Ritabrata Chakraborty) paper "On the effective transfer of knowledge from English to Hindi Wikipedia" has been accepted at the COLING 2025 (Industry Track) 🌟! In this work, we propose a lightweight framework that enriches Hindi Wikipedia articles by extracting and adapting updated information from English sources, ensuring alignment with Wikipedia’s style and policies, and then translating the adapted content into Hindi to significantly improve the coverage and quality of these articles. Key Contributions include: 1️⃣ We propose a multistage approach to extract knowledge from biographical writings about popular figures, transform this text to adhere to the NPOV guidelines and incorporate it into low-resource Wikipedia articles, e.g., Hindi. 2️⃣ We manually curate 103 biographical writings relevant to corresponding Wikipedia articles as external knowledge sources, which can serve as a rich source of factual information 3️⃣ Our lightweight framework efficiently updates Hindi Wikipedia articles by adding coherent and new information. 📄 The preprint is available on ArXiv: https://lnkd.in/dQ2axSx4 We are thrilled to present this work at COLING 𝟮𝟬𝟮𝟱 and look forward to connecting with the AI research community 🌐🤝.

64

4 Comments

2024: The year AI stopped being exclusive As I return from an incredible vacation in India, I find myself reflecting on my customary New Year's post. The highlight of my trip was experiencing Kashmir for the first time - and words can barely capture its breathtaking beauty. Looking back, 2024 has been nothing short of transformative, especially in the realm of machine learning. The pace and scale of innovations we've witnessed this year have fundamentally reshaped the landscape of AI technology: - The Battle Intensified: We saw a new LLM war intensifying between the tech giants with surprisingly Facebook being the most generous with their open sourced models. Gemini getting up to the challenge towards ending of this year while Anthropic(Amazon) also giving some good competition to openAI. - Branding Helped CharGPT: ChatGPT became the “Kleenex” of AI. Everyone says ChatGPT even when they use Claude! Speaking of which, we got to add one more subscription either chatGPT or Claude into our endless list of subscriptions. - Integration took centre stage: Integration of LLM models is on everybody’s mind. Nobody(in applied ML at least) is talking about working on foundational LLMs as LLMs move far into the research category now with the costs and the efforts associated with training them. - New “fad” jobs being created: particularly the no-code generative AI engineer and prompt engineers. The biggest shift? LLMs have truly democratised ML/AI as I see a lot of SMEs are working on integrating machine learning into their systems. For my predictions going into the next year: - Integration remains king. We will further go the integration route with ML systems engineer being a coveted post in future - Ideas didn’t used to matter before as much as applicable ideas were far and few. This dynamic will take a u-turn. - We are going to see a lot of resumes with big and complex ML stuff while candidates will just demonstrate surface level ML understanding in interviews. - Small and medium sized industries continue entering the AI game as they start to automate mundane tasks and improve their backend systems. This Democratisation of AI would be coupled with higher productivity, more GPT subscriptions and as you could imagine significant downsizing in future as jobs get automated with AI. What are your thoughts on this? Let me know in the comments.

26

3 Comments

Hello LinkedIn!! Just wrapped up the "Efficiently Serving LLMs" course from Deeplearning.AI, This deep dive into optimizing large language models was eye-opening. We covered some seriously cool tech: KV caching: Speeding up those massive models Batching: Handling multiple requests like a boss Quantization: Shrinking models without losing the magic Low Rank Adaptation (LoRA): Fine-tuning without breaking the bank Multi-LoRA: Juggling multiple tasks with finesse The hands-on approach really helped solidify these concepts. I'm pumped to apply these techniques in real-world scenarios and push the boundaries of what's possible with LLMs. Big thanks to the Deeplearning.AI team for putting together such a comprehensive course. #AIOptimization #DeepLearning #LLMs

27

1 Comment

🚨 𝗘𝘃𝗲𝗿𝘆𝗼𝗻𝗲 𝗶𝘀 𝗮𝗽𝗽𝗹𝗮𝘂𝗱𝗶𝗻𝗴 Blinkit's 𝗮𝗺𝗯𝘂𝗹𝗮𝗻𝗰𝗲 𝘀𝗲𝗿𝘃𝗶𝗰𝗲, 𝗯𝘂𝘁 𝗵𝗲𝗿𝗲'𝘀 𝘄𝗵𝗮𝘁 𝘁𝗵𝗲𝘆'𝗿𝗲 𝗻𝗼𝘁 𝘁𝗮𝗹𝗸𝗶𝗻𝗴 𝗮𝗯𝗼𝘂𝘁... While everyone's focused on "𝟭𝟬-𝗺𝗶𝗻𝘂𝘁𝗲 𝗮𝗺𝗯𝘂𝗹𝗮𝗻𝗰𝗲𝘀", they're missing the bigger revolution happening here. Here's what most people don't realize: The real innovation isn't the speed it's the complete redesign of emergency response: 1. Traditional ambulances aren't slow (10-30 mins average varies case by case), but they waste critical minutes in the dispatch process itself:🤔 - "What's your location?" - "Can you describe the emergency?" - "Please stay on the line…" - "I cannot find your location" In the golden hour where every minute counts, this back & forth just spoils everything.😔 2. Blinkit's masterstroke🙌: - The same app you order groceries from - One-click - Location already shared - Ambulance dispatched instantly -Arrived at your location like that delivery partner But here's the genius part no one's talking about:🤫 • Real-time tracking of emergencies • Crowd Source the information of the accident on the road for immediate response • Immediate help for gig workers themselves in case of need They're not just launching ambulances they're building India's largest emergency response network. Think about it: When did a quick commerce company last think about a saving initiative instead of just minting revenue? What makes me proud is that this isn't just another profit-driven venture. They are focusing on affordability (current traditional ambulance costs can go up to ₹2,500 with surge) Whats your most used Quick commerce app do you wish to see an emergency button inside it? #Innovation #Healthcare #Technology 

65

5 Comments

🚗 Introducing the AI Smart Parking Assistant! 🚗 I'm excited to share a project that played a pivotal role in our recent win at the national-level AI competition hosted by Tech4Change, Intel Corporation, and Lodha Oakwood School. The judging panel included esteemed people like Ram Srinivasan. As the mentor for our innovative team, I guided my younger brother Vishwajeet Zende and his 6th-grade classmates Anwita, Ananya, and Ved in creating the AI Smart Parking Assistant. This IoT-based project leverages the power of Arduino UNO, IR sensors, and a Python GUI to provide real-time monitoring and intelligent recommendations for parking spots. Here’s a brief overview: 🌟 Features: 📡 Real-time Monitoring: Up to 10 parking spots are continuously monitored. 🖥️ User-Friendly GUI: Easily visualize the availability of parking spots. 🤖 Intelligent Recommendations: Our system suggests the optimal parking spot for drivers. 🔌 Reliable Detection: An Arduino-based sensor network ensures accurate spot detection. 🔧 Customizable: The system can be scaled for various parking lot sizes. 💻 Technical Highlights: Hardware: Arduino Uno, IR sensors, and breadboards. Software: Python for the GUI, with libraries such as pyserial, tkinter, and Pillow. Installation: Simple setup with detailed documentation available on GitHub. This project not only showcased our technical skills but also highlighted the importance of real-world applications of AI and IoT in solving everyday problems. Our Smart Parking Assistant aims to optimize parking space utilization, reduce time spent searching for parking, and ultimately contribute to smarter, more sustainable cities. I’m proud of what our team achieved and excited about the potential impact of our project. Check out the working video of our AI Smart Parking Assistant below! For more details, visit our GitHub repository: [https://lnkd.in/gsbb8JaG] #AI #IoT #SmartParking #Innovation #Tech4Change #SustainableCities #Arduino #Python #Mentorship #YouthInTech

41

In my college hackathon, in 2017- I conceptualized something called as kanihastra. ( p.s: the name might sound cringey) You can assume it like a pendant with an IOT device. My business model was pretty simple- the pendant can be customized based on your requirements but the IOT will remain constant across. So what will the IOT do? - It was auto programmed to give a high alert to the nearby police station & the Guardian - with the location coordinates and starts recording the voice from then on - It only starts these actions when there's a trigger, here we programmed the trigger as a strong hit ( ~65-70 N) and or a strong scream from the concerned owners voice ( which can be synced with AI Alexa or Siri or custom built) While there was a big green signal for that idea back then, the hackathon team decided to reject us, citing a parking IOT solution as the best idea of the day. I feel pity on them. Today - I'm not in a position to execute this, but I can help and guide students who are in pre final and final year of CS, IT, ECE and Mech and make this happen. Two things why you need to do this now! 1- You can't just preach and teach women safety, we're used to it from early days. Act on it- our pendant might be sold at just 0.75 the cost of a pepper spray bottle 2- It has a business model with a social cause! You get paid for what you do. So that you'll have that satisfaction. I can connect you with the best mentors and can even help you get some funds to kick start. Recently saw a post from my dear akka Arasi Arul & thought of - why should we hesitate to bring this to limelight! Today, women's safety is at stake! We're compromising that easily. On the other hand, we shouldn't allow them to fall into fear, by up selling our safety ideologies. I've seen so many Singapengal in my life, starting from my mom, sisters and all my fellow entrepreneur Sago's Arasi Arul Janani Porkodi Archana Stalin etc etc They've broken the taboo and are marching towards the win!!! Let's make India a safe place to all our dear mom and sisters like how we made it for our own family.. Note: The commercial angle to this idea is much needed because it's not the fear we sell! it's the shame of every men in the country, that's what we actually sell & that has a production cost! Until or unless someone comes forward to support that, the team which is doing this needs a financial support to run. That's why the commercial angle comes into play. Please don't start criticizing that for your own sake p.s: This post isn't for men who still complain the dressing style of women is all the reason behind the happening- you guys pls stf & leave 🤞

23

7 Comments

The company's search boss, Prabhakar Raghavan, told employees at an all-hands meeting in March that Google plans to build teams closer to users in key markets, including India and Brazil, where labor is cheaper than in the U.S. Seems like SkillsFuture is leveling up in India and Brazil; also means the Commoditization of typical "Core" roles could be performed beyond the US, with much lower cost. Now, my question, what can our next-generation Singaporean fresh graduates do, than those countries mentioned yet to? Also goes to many staff with the business-as-usual SG&A roles, how to keep them from moving into Thailand and Malaysia?

52

23 Comments

Just stumbled upon this article. For those who are in sophomore year of their graduation and interested in ML and Gen AI. #AI #ML #GENAI

9

The Melodies of an Image: Exploring Music Recommendations Based on an Image's Content and Context Excited to announce our latest paper, "The Melodies of an Image: Exploring Music Recommendations Based on an Image's Content and Context," co-authored with Ishmeet Kaur. Presented at the 17th ACM Conference on Recommender Systems (RecSys), this research delves into a novel approach to music recommendation systems. By intertwining image content and contextual cues, we aim to provide more tailored and emotionally resonant music suggestions for users. 🎧 What's New? Traditional music recommendation systems often hinge on user behavior or metadata. However, our method goes beyond by analyzing visual and contextual elements of images to curate music that aligns with both the mood and setting of the image. From serene landscapes to vibrant concerts, or engaging social media posts, our research illustrates how this fusion enhances user experience. 🔍 Key Highlights: - Multimodal approach: Utilizing visual features (colors, textures, objects) and contextual data (captions, platform, interactions) for music suggestions. - Enhanced personalization: Matching music to the emotional essence and content of images. - Real-world applications: Envision tailored playlists based on shared images, or social platforms offering bespoke music recommendations! 💡 Why does this matter? Our findings hold promise for sectors like social media, photo-sharing apps, and music streaming services. By harmonizing image and music realms, deeper user engagement and enriched experiences can be crafted. 📄 Read the full paper here: https://lnkd.in/gQ6PNeji Grateful for the opportunity to share this at RecSys Conference, I look forward to the potential this research brings. Share your thoughts or queries in the comments—I value your feedback! 🙌 #Research #MusicRecommendation #AI #MachineLearning #ImageAnalysis #ContextualIntelligence #Personalization #MusicTech #ArtificialIntelligence #Multimodal #Innovation #ConferenceName

36

The Dilemma of Specialized Engineering Degrees: Are Students Really Job-Ready? Specialized programs like PGPs in AI and ML courses are booming in India, promising to prepare students for future jobs. But is the job market ready for them, especially outside startups? Prof. V Ramgopal Rao, Vice-Chancellor of Birla Institute of Technology and Science, Pilani, raises a red flag. He warns that these fancy titles might limit students’ prospects in government jobs or traditional master’s programs. While these programs are government-approved, the rigid eligibility criteria pose significant challenges, leaving students with fewer options. Is this the right path for India’s future engineers?

20

1 Comment

I wonder why high-impact research papers on AI are not published by Indian institutes but mainly by the US or China. If you're an alumnus of an Indian tech university or a researcher (either in India or abroad), I'd love to know about your perspective on this matter. #research #ai #artificialintelligence #machinelearning #datascience #ml #softwareengineering #iit #iiit #tech #technology

2

🌟Celebrating successful Google Summer of Code projects🌟Abhay Lejith, Armaghan Shakir, Felix Hirwa Nshuti, Pranav Prajapati, Xinyu Wu 🌟 We are excited to share that another round of highly successful summer projects have concluded at sktime, with generous funding by Google Summer of Code. It has been extremely rewarding for our mentors to witness our brilliant mentees working with the community, bringing fresh perspectives, energy, and dedication to our mission. Their contributions have propelled our community to new heights, and we could not be more proud of their achievements. Abhay Lejith - categorical feature support for forecasting https://lnkd.in/ez9reQHe Armaghan Shakir - integration with pytorch, huggingface https://lnkd.in/e_e8G899 Felix Hirwa Nshuti - scaling backends, pytorch and darts integrations https://lnkd.in/e2xbekbK Pranav Prajapati - polars integration, foundation models, transformers https://lnkd.in/evjUPz4K Xinyu Wu - global forecasting API, pytorch-forecasting integration https://lnkd.in/eb8c8PN5 On a more somber note, we are also sad that we were unable to bring everyone together in a single physical location to celebrate our successes - it has been extremely upsetting to witness which disadvantages, indignities and corruption even the most highly sought-after technical experts are forced to face, for instance as part of visa processes, even for a few days of conference travel, just because of name, skin colour, or country of birth. It is in moments like this where we again realize the importance of our mission (https://lnkd.in/ePNQmFxG), and the need for suitable operational and governance frameworks to reverse these ubiquitous mechanisms of exclusion. Yet the path is bright: we are more than happy that our mentees have decided to become integral members of the sktime community, and we look forward to building an open, fair, and collaborative future together. #sktime #gsoc #opensource #python #timeseries #ai #ml #datascience

73

1 Comment

iOS vs Android really?? You might have read lot of articles on iOS vs Android prices differing on ride booking apps like Uber, Ola or Rapido. But is it truly "just" the operating system at play? Not exactly, There's a lot more happening under the hood, lets get into the hidden tech behind these pricing strategies: - Dynamic Pricing Based on User Behaviour and Historical data These platforms rely on user behaviour patterns to adjust pricing dynamically. Once they identify a "regular" user, the fares can be inflated. Similarly if a user has not booked a ride for a long time or seems to be a new user, they might get "fancy" discounts. - Platform-Specific Pricing Configuration(𝐘𝐨𝐮 𝐦𝐮𝐬𝐭 𝐡𝐚𝐯𝐞 𝐠𝐮𝐞𝐬𝐬𝐞𝐝 𝐭𝐡𝐢𝐬 𝐨𝐧𝐞!!) When the app makes requests to the server (for ride pricing or availability), it sends headers that include the user-agent string eg. "Android/10" or "iPhone; CPU iPhone OS 14_4 like Mac OS X". The service might have different base fares, surge pricing algorithms, or discounts for Android and iOS users. - A/B Testing and User Segmentation Companies may use A/B testing or user segmentation to experiment with different pricing for Android and iOS users. Uber might randomly assign certain users to either a test group or control group where they experience different prices, and track how these changes affect ride patterns, customer satisfaction. - Platform-Specific Discounts or Promotions Ride booking apps may run exclusive promotions for users on one platform to encourage adoption or usage. For example, Rapido might offer discounts for iOS users to promote the app's presence in the Apple ecosystem, especially during special events like product launches or holidays. Similarly, they could incentivise Android users with discounts to increase market share, especially in regions where Android dominates. - Geographic Targeting and Regional Pricing Variations Surge pricing could vary between platforms in different geographic locations based on market dynamics. Also, pricing for Android might be lower in areas where Android has a larger user base and higher in regions where iOS leads. Next time you book a cab, auto or bike do remember these apps have a lot of Machine learning and AI based targeting algorithms which they use to help maximise revenue and optimise user retention. #softwareengineering #uber #ola #rapido #pricing #ios #android

18

2 Comments

Graduating From A Private College to Landing A Dream Job At GOOGLE | Saurav's Journey, Study Plan & DSA Tips Saurav’s story shows that where you start doesn’t have to determine where you end up. Coming from a tier 3 college, he knew the challenges ahead. But instead of letting that hold him back, he focused on learning, staying updated with the latest industry trends, and sharpening his coding skills. There were tough times—long nights, self-doubt, and rejections. But Saurav didn’t give up. He kept pushing forward, using every setback as motivation. And now, he’s landed a job at Google, one of the top companies in the world. Watch the Full Episode- Link in Comments! Saurav’s journey isn’t just about getting a dream job. It’s about never giving up, always growing, and believing that you can achieve anything if you keep working at it. If you ever feel like your background is holding you back, remember Saurav’s story. It’s not about where you start, but where you’re headed. With determination and a willingness to learn, you can reach your goals, no matter how big they are. Here’s to breaking limits, proving people wrong, and creating your own success! #coding #google #gfg #geeksforgeeks #softwareengineer #coder #jobs #careers #btech #hiring #mtech #corporate Saurav Prateek | Sachin Sharma | Sandeep Jain

73

5 Comments