Deepak Mittal🤖

Recently, PW (Physics Wallah) launched Aryabhata 1.0, an AI model built using something called an SLM, which stands for Small Language Model. Unlike big models like GPT-4 or Gemini (called LLMs – Large Language Models) that need a lot of data, power, and memory to run, SLMs are smaller, faster, and easier to use. They are made for specific tasks and give quick answers without using too many resources. This makes them a good fit for education, where speed and accuracy are more important than handling every type of question. While LLMs try to do everything, SLMs are focused — perfect for things like doubt-solving, tutoring, and exam help. Note: A lot of news has surfaced claiming that it is just a wrapper on top of the Qwen model. However,this post is solely focused on how LLM and SLM models differ.

730

8 Comments

Lately there hasn’t been a dull moment in the R-world when it comes to working with LLMs. And as I said in one of my earlier posts: packages emerge faster than you can type 𝚒𝚗𝚜𝚝𝚊𝚕𝚕.𝚙𝚊𝚌𝚔𝚊𝚐𝚎𝚜() 😬. But the same can be said for updates to those packages! In today’s spotlight: 𝗲𝗹𝗹𝗺𝗲𝗿 version 0.3.0 🐘 If you’re not familiar with 𝗲𝗹𝗹𝗺𝗲𝗿 yet, here’s a short summary: 𝗲𝗹𝗹𝗺𝗲𝗿 helps you to call LLM APIs from R in a consistent way, and of course, this opens up a world of possibilities. It’s proving to be a handy back-end for various other LLM-related packages. You could say it’s the workhorse in R’s “LLMverse”. So what’s new? 🔹a handy new function called 𝚌𝚑𝚊𝚝() which can be used instead of specific provider functions like 𝚌𝚑𝚊𝚝_𝚊𝚗𝚝𝚑𝚛𝚘𝚙𝚑𝚒𝚌() or 𝚌𝚑𝚊𝚝_𝚘𝚙𝚎𝚗𝚊𝚒(). It simplifies the chat interface: you now just specify the provider and model in a single string. 🔹defining tools for function calling is also simplified. A tool is just an extra ability an LLM can use. You’re basically saying: “here’s a function you’re allowed to use, this is what it’s called, what it does, and what kind of input it needs”. Specifying that is more intuitive now. Note that this change is breaking. 🔹we hate big unexpected bills so we like to track costs 💸. The latest version of ellmer contains another pricing data source so there’s info for more providers/models. Simply call the 𝚐𝚎𝚝_𝚌𝚘𝚜𝚝() method. PS: 𝗲𝗹𝗹𝗺𝗲𝗿 and 𝗰𝗵𝗮𝘁𝗹𝗮𝘀 (the Python equivalent of 𝗲𝗹𝗹𝗺𝗲𝗿) are closely aligned. I expect (or hope) 𝗰𝗵𝗮𝘁𝗹𝗮𝘀 will follow the simplified chat interface too. Tool spec is already intuitive there and the cost thing is taken care of in the dev version.

165

18 Comments

Some people crack AI/ML roles at Google, Microsoft, Apple on their very first attempt. But nobody tells you this secret 👇 It’s not luck. It’s a method — and anyone who follows it with focus and consistency can do the same. 🔹 The “Math & ML First” Principle Before touching deep learning frameworks, they master the core: • Linear Algebra & Calculus • Probability & Statistics • ML Algorithms: SVM, KNN, Decision Trees • Optimization (SGD, Regularization) Because interviewers don’t just ask you to use models — they ask why and how they work. 🔹 The “Concepts Over Code” Rule They don’t rush to write 10,000 lines of TensorFlow or PyTorch. They first understand bias-variance tradeoff, overfitting, feature engineering, and model evaluation. Then they code. 🔹 The “Projects That Matter” Approach Instead of copying 10 Kaggle notebooks, they build 2 solid end-to-end projects: • With real datasets • Using data cleaning + ML pipelines • APIs, dashboards, or model deployment That’s what hiring teams remember. 🔹 The “Paper + GitHub” Combo They read 2–3 core ML/DL research papers — and replicate or extend them on GitHub. This shows real depth. Not just course knowledge. 🔹 The “Mock & Explain” Mindset They practice mock interviews — not just solving, but explaining: • Why they chose this model • Why accuracy dropped • Why this feature matters Because AI interviews test your thought process, not just code. If you're serious about cracking top ML/AI roles — Stop chasing random tutorials. Start following a proven strategy that works. Say “MLMETHOD” and I’ll share the exact roadmap followed by folks who made it to Google Research, Razorpay AI, Microsoft Applied Sciences, and more. Someday, someone will say: “You just got lucky.” But you’ll know — you followed the system nobody talks about. Follow Raunak Yadush for hiring updates and referrals — whether you're a fresher or switching jobs. DM me. Always happy to help. Instagram: https://lnkd.in/gm_g6Q3Z Telegram Group: https://t.me/raunakyadush #MachineLearning #AIJobs #MLInterviewPrep #TechCareers #SDEPrep

6,194

49 Comments

Introducing DynaRoute - intelligent model routing for efficient LLM usage At Vizuara, we (Pritam Kudale, Pranavodhayam P R, Raj, Rajat) have spent the last few months studying how teams use large language models in production. What we found was simple but eye-opening. Most teams rely on one large and expensive model for every kind of query, whether it is a quick factual question or a complex reasoning task. This works, but it wastes a lot of money and computing power. That is what led us to build DynaRoute (https://lnkd.in/dGSx2KRq), an intelligent routing layer that automatically analyzes your queries and sends them to the most suitable model based on complexity, cost, and performance. You do not have to change your application or rebuild your backend. You can continue using your familiar OpenAI-style API, just by switching the base URL and key. From that point, DynaRoute handles everything intelligently behind the scenes. In our internal benchmarks, DynaRoute delivered remarkable results. Real-world employee questions achieved up to 92 percent cost savings, while code-related queries saw around 74 percent savings. All of this happened while maintaining nearly the same quality and accuracy as top-tier models. For teams that are running customer support systems, coding assistants, agents, or RAG-based pipelines, DynaRoute can reduce your inference bills drastically while preserving the quality your users expect. You can think of it as a smart, adaptive router that continuously learns how to choose between models like GPT-4.1-nano, Gemini 2.5 Flash, and Claude Sonnet, depending on the query. We also provide API access so you can integrate DynaRoute directly into your own platform or agent framework. The API is OpenAI-compatible and can be configured with just a few lines of code. Here are our current plans: Free Tier: ₹0 per month, includes up to 50,000 tokens, OpenAI-compatible endpoint, and access to premium models. Plus Tier: ₹888.30 per month, includes up to 3 million tokens, unlimited requests, access to premium models, and priority support. Enterprise Tier: completely custom implementation for companies that need dedicated support, detailed analytics, and on-premise deployment. We also provide custom routing models for enterprise-level optimization. We are currently offering Enterprise Plan API access to selected partners who wish to integrate DynaRoute into their platforms. If you want to optimize the cost and performance of your LLM-based applications without compromising output quality, give DynaRoute a try. 👉 Learn more and get your API key at https://lnkd.in/gTrs9NWA

81

3 Comments

🚨 India’s consumption puzzle: Why do two official datasets tell different stories about how we spend? PFCE (top-down; macro-consumption) vs. HCES (bottom-up; micro-consumption) 🔹 Gap has widened from 5% in 1972–73 to ~45% today. 🔹 Free/subsidised services (education, healthcare, meals) inflate PFCE but barely appear in HCES 🔹 Food share in household budgets is falling; non-food (transport, healthcare, rent, education) dominates This divergence isn’t just statistical—it affects how we measure poverty, prosperity, and the impact of policy. 📊 Latest piece by Ashish Kumar and Payal Seth from Pahlé India Foundation (PIF) in businessline unpacks the data divide and calls for urgent reforms (https://lnkd.in/dJWjyCQe)

179

8 Comments

Python Curriculum at Deerwalk Schools – Preparing Kids for the AI Era At Deerwalk, we have been teaching coding since Grade 1. Now, we are making it more structured from Grade 3 onward and putting stronger focus on Python—the world’s #1 language for AI, data science, and automation. Why Python? Because coding is today’s literacy. Just as students must learn to read, write, and calculate, they must also learn to code. Python is simple enough for a 10-year-old to start, but powerful enough to run Google, NASA, and ChatGPT. By the time Deerwalk students reach Grade 10, they will be fluent coders, ready to apply Python in real-world projects. We are sharing our roadmap below not just for Deerwalk parents and teachers, but also for other schools in Nepal who want to follow Deerwalk’s lead. Python Roadmap at Deerwalk Schools Grade 3–4: Logic & Fun with Code Block-based coding (Scratch, Blockly) Transition to Python with Turtle graphics Projects: Animated stories, drawing a house with Python Grade 5: First Steps in Python Variables, input/output, Turtle shapes Projects: Guess-the-number game, digital dice roller Grade 6: Building Foundations Loops, conditionals, random numbers Lists and simple data storage Projects: Rock-Paper-Scissors, simple chatbot Grade 7: Expanding Skills Functions and modular code Dictionaries, file handling basics Projects: Word counter, Hangman game Grade 8: Problem Solving Search & sort algorithms Error handling, debugging Projects: Student grade tracker, text adventure game Grade 9: Applied Python OOP basics Data visualization with Matplotlib APIs and real-world data Projects: Weather app, expense tracker Grade 10: Python Fluency Web basics (Flask) Data analysis with Pandas Intro to AI/ML (Scikit-learn) Projects: Sentiment analyzer, quiz web app Grade 11: Python for Innovation Advanced OOP & design patterns Machine learning projects Robotics with Python (Raspberry Pi/Arduino) Projects: Handwriting recognition, smart home simulation Grade 12: Python for the Future Deep Learning (TensorFlow/PyTorch) Capstone projects solving real-world problems Projects: AI chatbot, recommendation engine Why This Matters By following this roadmap, Deerwalk students will graduate with 10 years of structured coding experience, fluency in Python, and confidence to build AI-powered applications before leaving high school. This is how Deerwalk is preparing the next generation. We invite other schools to join us in giving kids the real skills of the future.

88

5 Comments

ChatGPT's new $5 "ChatGPT Go" plan: an opportunity or a death sentence for Indian AI startups? Yesterday, OpenAI released a $5/month tier specifically for price-conscious markets like India, Nigeria, and Brazil. One coffee per month now gets you GPT-5 with a good-message daily limit. For millions of students, freelancers, and nascent founders, this is a no-brainer. But for Indian SaaS and AI-product firms that have been taking ₹499–₹999 for comparative wrappers, the arithmetic just got merciless. Here's why this is more than simply another price shift: 1️⃣ Anchor-price reset As the world standard drops to ₹400 ($5), all such local "AI helper" or "co-pilot" now looks too pricey-unless it provides transparent, defendable value on top of GPT-5. 2️⃣ Distribution chokehold OpenAI already owns the user relationship. A $5 funnel will vacuum up top-of-funnel traffic that Indian startups have been paying Google and Meta dearly to acquire. 3️⃣ Cash-burn asymmetry OpenAI can subsidize $5 users; a pre-Series A startup cannot. The longer the subsidy lasts, the wider the moat becomes. 4️⃣ The silver lining OpenAI’s $5 plan only covers low-volume, general-purpose use. Edge cases-Indian languages, hyper-local compliance, offline-first workflows, private-cloud deployments-remain wide open. So what do Indian founders need to do, pronto? • Go vertical, not horizontal: Fix ICD-10 coding for Indian hospitals, GST invoice parsing for CA companies, or KYC audio verification in Hinglish-domains OpenAI can never optimize against. • Build moats with data: If your training data is 100k exclusive PDFs from Indian courts, no vanilla model can match that overnight. • Bundle services, not tokens: Layer on BPO, human-in-the-loop review, or regulatory filing-things a $5 chatbot can’t replace. • Price on ROI, not tokens: A recruiter who saves 10 man-hours per hire will happily pay ₹2,000; the sticker price war becomes irrelevant. The takeaway: Compete on atoms, not bits. The moment your value is measured purely in “tokens per rupee,” you’ve already lost to OpenAI. Agree? Disagree? #BuildInIndia #AIStartups #OpenAI

32

"I am studying so hard for AI / ML interview, but can't crack it" - if you are too in the same zone, come out of traditional preparation loop. Here's what you actually need to crack AI / ML / GenAI interviews in 2025: ✅ 1. Solve Coding Problems - every single day Most people get rejected here without moving further. Companies will ask one DSA for sure: - Arrays, Strings, Hashmaps - Sliding window, Binary Search, Two pointers - Matrix traversal, Graphs, DP (optional but good to have) Study material: LeetCode 150 / Striver's DSA sheet / GFG SDE sheet Set a target: 2 DSA questions/day = 60/month = a real confidence booster! ✅ 2. Build Projects That Teach You and Impress Recruiters Skip toy datasets and MNIST clones. Instead, build real-world GenAI and ML projects like: - A Resume Screening Bot using LLMs - Fake News Detector from live RSS feeds - Multi-label Tag Predictor for articles / e-commerce products - Semantic Search App using Vector DBs Build → Deploy → Document → Share (Linkedin / Reddit / X) That’s the cycle, make sure you know end-to-end of a project. ✅ 3. Master ML System Design — It’s Asked in Every Interview Now This is a reminder that System Design is non-negotiable. Some case studies include: - Design a system that summarizes customer support tickets in real time - Walk me through how you'd deploy an LLM-based assistant. Train yourself to: - Break down problems into small components - Identify edge cases, build evaluation metric - Choose trade-offs (accuracy vs latency) - Design for scale and data drift Tip: Maintain a personal repo of 10+ system design case studies with diagrams. If you maintain this healthy study routine for 2-3 months rigorously, you will be 100% ready for interview with confidence. ______ If you need guidance and mentorship, book a 1:1 call with me. Click 'Book and Appointment' above.

1,088

59 Comments