Darshita Chaturvedi

New (sponsored) post - we go through what Innovaccer actually does in an interview with Abhinav Shashank - The process by which unstructured data becomes structured - Their customers and example use cases for each of them (including screenshots of the actual product itself) - How they think about which applications they should build on top of the data, when they should partner, and how they think about the EHRs native applications you can read the whole thing here: https://lnkd.in/gHfndu72

263

29 Comments

IIT AI weekend discussion. It was a fun Sunday morning discussion on AI use cases from across industries. Thanks everyone for joining! Here are some takeaways: 💡 AI Agents and Automation Engineering: One participant built a system that automates software engineering workflows using GPT for PRD creation, database schema development, backlog creation, and sprint planning. End to end automation is now possible. Headcount reduction: AI agents are significantly disrupting back-office functions across industries. 10-person teams could potentially be reduced to 2-3 person team. Some participants expect AI agents to report to them by the end of 2025. Design: Figma Make (using Claude) is popular for rapid product prototyping. It becomes easy to convert basic wireframes to full responsive designs which reduces the need for large design system teams. Mobile-responsive designs can now be generated in minutes. Document processing: Companies are deploying orchestration layers to manage and monitor agents. Dense text can easily be converted to visual frameworks. Use cases include document generation, summarization, and specialized task automation. 💡 AI Use Cases Across Industries Insurance: Automating workflows, compliance processes, and property valuation. Robotics and Hardware: Accelerating physical AI development and integrating AI with engineering tools. Cybersecurity: Building AI agents for ethical hacking and security automation. Edtech: Utilizing low-code/no-code tools for rapid product development. Enterprise: AI adoption accelerating in HR, legal, procurement, and customer success functions. 💡 Limitations of AI agents Several participants mentioned AI tools “falling apart after a few steps” or requiring significant human guidance. Agents perform well for initial tasks but often fail after multiple steps, requiring human intervention to complete complex workflows. Agents struggle to handle specialized or nuanced tasks that require deep domain expertise. Fine-tuning for specific industries or workflows is still a manual and resource-intensive process. Handling regulatory compliance and preventing hallucinations remains challenging, particularly in regulated industries like insurance. 💡 Disruption and Moats in AI Differentiation is increasingly difficult with foundation model vendors, cloud hyperscalers, and others adding similar capabilities. Traditional technology moats are eroding with the commoditization of AI tools. Distribution and vertical domain expertise are becoming key differentiators. Personalized and user-specific solutions offer potential long-term advantages. Roll-up strategies (e.g., acquiring smaller companies for data and workflows) are gaining traction.   Trust, UX, and virality are critical for success; niche-focused tools can thrive. Rapid scaling and adoption are possible, but churn and delivery quality remain challenges. #IIT #ai #founders

147

8 Comments

OpenAI just launched chatgpt go in india at ₹399/month 🙇🏻‍♂️ you get 10x higher message limits, 10x more image generations, 10x more file uploads, and double the memory compared to the free plan. plus, payments via UPI, no more USD conversion headaches. india’s the first country to get this tier, aimed at students, freelancers, and anyone who found the free limits too restrictive but didn’t want to pay for plus. the go plan bridges that gap, making advanced AI features way more accessible. feels like a smart move: affordable, local payments, and more power for everyday users. curious to see how this shapes ai adoption here 🙌

157

9 Comments

How Baaz Bikes is Leveraging AI Across Their Operations Recently I asked one of our portfolio companies, Baaz Bikes co founder Anubhav Sharma, about their AI adoption journey. The breadth and depth of implementation is impressive – here's what they're doing: Customer Intelligence & Retention Customer DNA & Churn Analysis Using ML classification models (Logistic Regression, XGBoost, Neural Nets) to predict churn, segment riders, and trigger personalized retention campaigns – directly optimizing marketing spend and improving customer lifetime value. Operational Excellence Swap Demand Prediction Time-series models (ARIMA, LSTM, Prophet) combined with AutoML platforms (Amazon SageMaker, Vertex AI) to forecast battery demand by location and time. This optimizes inventory allocation and minimizes downtime – critical for a battery-swapping business. Predictive Maintenance IoT sensors capturing motor, battery, and environmental data feed ML/Deep Learning models to predict breakdowns before they happen. The result: proactive maintenance, extended asset life, and reduced repair costs. Support & Communication Multilingual Customer & Ops Support Leveraging Rakuten Gist Genie, Rakuten Shoptimiser, and Conversational ML Models (NLP) for multilingual rider support, spell/grammar correction for staff, and SOP-driven training. This enables faster resolution across India's diverse linguistic landscape. Enhanced Collaboration Slack/Teams AI copilots, meeting summarizers (Otter, Fireflies), and multilingual bots to translate tickets, auto-summarize meetings, and surface relevant documentation in real-time. Analytics & Insights AI Business Analytics & RCA Data warehouses (Snowflake, BigQuery, Redshift) combined with BI tools (Looker, Power BI, Tableau) and anomaly detection using Rakuten AI for real-time monitoring, root cause analysis, and actionable insights that drive profitability. Data Pipeline Optimization AI-powered data quality monitors, Rakuten SixthSense, and ETL optimization tools ensure clean, reliable data for ML models and faster analytics. Developer Productivity Code Assistance & Quality GitHub Copilot, AI linting tools, and auto test generators accelerate development, automate reviews, reduce bugs, and improve onboarding for new engineers. What stands out: This isn't just AI for the sake of AI. Every use case ties directly to unit economics, customer experience, or operational efficiency. They're building AI into the DNA of how they operate. AI adoption doesn't have to be all-or-nothing. Baaz Bikes shows how to systematically deploy AI across the value chain – from customer-facing applications to internal operations – while using best-in-class tools for each specific use case. #AI #MachineLearning #StartupOperations #EVRevolution #Portfolio #BaazBikes Kalaari Capital

72

Quick commerce might create new rails for fashion in India. But AI is about to rewrite the stack. It won’t just improve margins or automate workflows. It will reshape how demand is created, what gets made, and how we buy. Here’s my prediction: 1. Search becomes intent-led Nobody wants to scroll through 400 SKUs. AI will learn your taste, body, budget, event, and mood, and surface five things that just work. Think: Spotify-style discovery, but for clothes. Discovery becomes contextual, not chaotic. We’re already seeing this in early interfaces like Perplexity’s shopping copilots. 2. Assortments get micro-targeted Massive catalogs are a liability. AI lets brands adapt SKUs dynamically, by user, region, season, even returns history. Shein scaled fast fashion through supply speed, but never cracked fit. Newme is flipping the model by doing weekly drops of 10–15 SKUs based on real-time feedback As merchandising behaves like content, inventory becomes a live system. 3. Returns are engineered out Returns were the biggest margin killer. Now they’re a solvable product problem through predictive sizing + fit-tech + try-at-home delivery. Zalando and H&M are already running fit-tech integrations + virtual try-ons at scale. Fit-tech will become table stakes. 4. Supply chains go real-time From design to drop to replenish to clear. AI enables live demand forecasting, smarter markdowns and faster reaction cycles. Urbanic, Zara, and Myntra are tightening feedback loops using browsing + returns + trend signals Fashion will respond to signals, not seasons and less dead stock will lead to better margins. 5. Shopping shifts from search to recommendation Shopping will shift from browsing to context-driven nudges. AI copilots will shop with you, not for you. Voice-first agents are already live. AI doesn’t just improve conversion: it changes the loop. The next generation of fashion brands will scale through personalization, fit precision, intelligent curation, and habit-forming UX Fashion will live at the intersection of fast-moving infrastructure and intelligent systems. This wont change how we buy. It will change what gets made.

123

20 Comments

Am noticing a tendency to go after low-hanging use cases amongst Indian AI founders. These include copy-cats, inserting “agents” into SaaS-era use cases without fundamentally re-imagining things, use cases that are prone to getting some early traction but lacking eventual depth etc. It’s also striking that in almost all cases, founders don’t have a particularly strong fit with the chosen market and therefore, lack a right-to-win against US competitors. I would suggest deeply thinking through upfront which problem statements to solve for. The whole point of this transformation is to reimagine things from first principles. Putting a fresh lens on underserved markets. Or unlocking new value that exponentially expands existing markets. Equally important is to layer in your own founder-market-fit in an intellectually honest way. Why is your team positioned to win against global competitors? What is the rabbit hole that you have gone down further than likely anyone else? What is the aspect of the problem that you just uniquely “get” more than 99% people out there? This thought-space is where great companies get initiated.

84

24 Comments

🚀 Introducing traversaal.ai's newest jewel Mantra-14B: A bilingual Hindi-English LLM that punches well above its weight — outperforming models twice its size. Mantra-14B is designed to bridge the language gap in real-world applications across South Asia and beyond. The model comes with 3% improvement in mean benchmark scores for both English and Hindi benchmarks for models twice the size. 🔍 Why Mantra Matters: - Outperforms larger models like Aya-32B, Krutrim-12B, and Nemotron-Hindi-4B - Excels in code generation, reasoning, and Hindi comprehension - Demonstrates lower bias in multiple-choice tasks - Strong domain adaptability and localized knowledge Released under the Apache 2.0 license Mantra sets a new bar for culturally aligned, multilingual LLMs that are efficient, powerful, and truly inclusive. Whether you're building for Bharat, training custom agents, or deploying open-source AI—Mantra gives you the foundation to go further. 🧪 Model Card: https://lnkd.in/d9yZKzs9 🚀 Colab Demo: https://lnkd.in/dauvyBJF 🌐 HF Space: https://lnkd.in/dbMDhzsk 📄 Paper: https://lnkd.in/dvEUNZeX Congratulations to traversaal.ai's entire team on such an amazing lauch!

65

2 Comments