Rishabh G.

India’s Biggest AI Access Revolution Just Began Google has partnered with Jio to bring its premium Gemini 2.5 Pro AI model to millions of users — completely free for 18 months. That’s ₹35,000+ worth of advanced AI tools now in everyone’s pocket. This isn’t just another offer. It’s a signal: AI is no longer a luxury—it’s infrastructure. When nearly half a billion users in India get direct access to Google’s best AI, the entire ecosystem shifts. Developers will experiment faster. Businesses will expect AI in every workflow. Startups and agencies must move beyond “using AI” to “delivering outcomes with AI.” As someone who’s worked on engineering leadership and automation for years, I see this as the moment where AI adoption becomes mainstream — not in theory, but in daily life. The question now is simple: How will you stand out in a world where everyone suddenly has access to the same AI power? I have registered for it, have you? #AI #GoogleGemini #RelianceJio #IndiaAI #AITransformation #EngineeringLeadership #DigitalIndia Full Article here: https://lnkd.in/gRMeCqrB

4

I am tired of seeing Indian engineering students graduate with a 9.5 GPA but zero production-ready skills. So, my team built a compiler for the academic noise. 🚀 As a student, you easily spend over 65% of your semester trapped in compliance: • Copying dense lab files manually. • Scrolling through bloated 200-page university slide decks. • Decoding unstructured, confusing syllabi. This is absolute cognitive overhead. We are prioritizing administrative compliance over actual engineering capability. That’s why my team (Sameeksha Prasad and team) built ACAMIX (Academic Efficiency Engine) for the Google Gemini Challenge. 🛠️ ACAMIX doesn’t change what your university teaches—it just compresses how fast you adapt to it. How it works: 1️⃣ Multimodal Ingestion: Drop any messy university textbook, PDF, or 5-year Past Year Papers (PYQs). 2️⃣ Gemini 1.5 Pro Compiler: It processes the unstructured data across a massive context window using strict typed JSON schemas to strip out the administrative fluff. 3️⃣ The Execution Bundle: You instantly get a "Precision Cheat Sheet" with Exam-Day Pitfall Warnings, plus a 90% ready structural outline for your labs via the Assignment Co-Pilot. 4️⃣ The Time-Flip: The dashboard calculates exactly how many hours you saved, and instantly converts that time into a hands-on technical roadmap (React 19, Python AI agents) inside a simulated sandbox terminal. We built the frontend using React 19 + Vite, styled it in an "Elegant Dark" bento layout with Tailwind CSS, and backed it with an Express.js (TypeScript) server. Stop wasting weeks on academic boilerplate. It's time to automate the grind and focus entirely on building real-world software. Check out our architecture and dashboard layout below! 👇 #BuildInPublic #GeminiAI #GoogleAIStudio #React19 #WebDevelopment #StudentInnovation #ProductDesign #StartupIndia

4

A research fellow at CERN doesn't usually end up building EdTech in India. Here's how that happened. At CERN, I worked alongside some of the sharpest minds on the planet. Physicists, engineers, mathematicians from 100+ countries. The thing that struck me wasn't the science. It was the methodology. At CERN, you don't guess. You don't assume. You measure. Every hypothesis gets tested against data. Every result gets validated before anyone believes it. Then I came back to India and watched companies hire people based on a 30-minute interview and a piece of paper called a resume. No measurement. No diagnostics. No data. Just gut feeling dressed up as "culture fit." I spent the next decade at Microsoft, Oracle, and EY watching this play out at scale. Now I'm building the diagnostic layer that should have existed all along. Pankh.AI doesn't replace interviews. It replaces guessing. What's one thing your hiring process measures that actually predicts job performance? #AIinEducation #SkillGap #FutureOfWork #HiringIntelligence #PankhAI

13

Building One of the Largest Multilingual AI Datasets for Indian Languages: Tonight our FoundryAI Labs Synth Engine v2 is running a massive dataset generation job. Current pipeline: • 19 academic subjects (Math, Biology, Chemistry, Geography, History, Civics, etc.) • 95,000 instruction seeds generated through self-seeding • Expansion factor: 6× via synthetic augmentation • 12 Indian languages being translated and validated Projected Output 95,000 seeds → 665,000 English instruction-answer pairs → Translated into 12 languages → ~7.98M multilingual samples After quality filtering (~85% pass rate): ~6.5–6.8 million high-quality dataset entries What this enables This dataset will power the training of BharatLLM, an open multilingual AI model focused on Indian languages. Goals: • AI tutoring for students (Grades 6–12) • Multilingual assistants for Indian users • Speech + text AI across major Indian languages • Open ecosystem for developers Tech Stack Behind the Pipeline • Synthetic Dataset Engine (Synth Engine v2) • Instruction seeding + augmentation • Multilingual translation pipeline • Automated quality filtering • JSONL dataset generation for LLM training The dataset will be used to train: BharatLLM (Mistral-7B + LoRA adapters) Final Dataset Size ~6.5M multilingual instruction samples ~1.5–2GB training dataset This will become one of the largest open Indian-language instruction datasets. Why this matters Most LLMs today are heavily English-centric. But India has 1.4B people and 20+ major languages. The future of AI must be multilingual by default. Next Steps • Complete dataset generation • Validate dataset quality • Train BharatLLM adapters • Open-source datasets + models Building the foundation for India’s multilingual AI ecosystem. #AI #LLM #OpenSourceAI #GenerativeAI #IndianAI #MachineLearning #DeepLearning #FoundryAILabs https://lnkd.in/e858kr4U

5

An Indian developer built Laya, the same class of system as Jev, and open-sourced it under Apache 2.0. Laya also answers questions in a single pass instead of generating tokens one by one. The repo shows 30.8 ms latency on a T4. On AG News, Laya's 95.3% surpasses Jev's 91%. But Jev wins Banking77: 87% vs Laya's 42.5%. These are not head-to-head tests and should be treated carefully. Still, the contrast is clear: one backed by $40M and closed, and another built in India and open-source for everyone.