Sidharth Pipriya

UPI now has an MDR charge — but here's why you shouldn't panic 🧵 There's a lot of confusion around this. Let's break it down: 1. It's not a Tax MDR (Merchant Discount Rate) doesn't go to the government. It's not part of the Consolidated Fund of India. This charge goes to the Payment Service Provider and the Bank behind it (e.g., PhonePe + its partner bank). 2. 95-96% of transactions are untouched MDR only applies to P2M (Person-to-Merchant) transactions above ₹2,000. As per NPCI data, ~96% of merchant transactions are either below this threshold or fall under the small-merchant exemption — so they're unaffected. 3. P2P remains completely FREE Sending money to a friend, rent, family transfers — all of this stays free. 4. Why the industry needed this UPI companies have been running in pure "Jan Seva" mode. Cross-sell revenue is negligible compared to daily transaction volume. No player in the industry has managed to make UPI profitable. A zero-revenue model can't sustain infra, fraud prevention, and uptime at global scale indefinitely. 5. How much will it actually cost P2M above ₹2,000 → 0.4% MDR, capped at ₹300 (the cap kicks in for transactions of ₹75,000 and above). Compare that to: Credit Card ~1.5-3%, Debit Card up to ~0.9-1% UPI is still the cheapest payment rail by far. 6. Small merchants are protected Street vendors and small businesses collecting up to ₹1 lakh/month via UPI QR (P2PM category) get zero MDR. A few more things worth adding: 📅 Effective from 15 October 2026 💰 Who gets the 0.4%? It's split between three players in the chain: (A) Issuer Bank (~40% share) — the bank where your account sits (e.g., your salary account bank that actually debits the money when you pay). (B) Merchant Acquirer (~30% share) — the bank/institution that onboards the merchant and settles the payment into the merchant's account (essentially the merchant's banking partner for accepting payments). (C) Partner Bank / UPI App (~20-10% split) — the UPI app you use (like PhonePe, GPay) and the bank technically powering that app's payment rails in the background. Think like: your bank moves the money out, the merchant's bank brings the money in, and the app in the middle facilitates the whole transaction — all three share the fee for keeping the pipes running. 🚫 No pass-through to consumers — Finance Ministry has explicitly said merchants cannot charge this to customers, & UPI apps are barred from adding any hidden platform fees. 🚆 Special categories: Railways, telecom, insurance, fuel get a flat ₹5 fee on transactions above ₹2,000. 📈 Capital markets (mutual funds, stockbroking) get a much lower 0.02% capped at ₹300. Bottom line: This is about making the ecosystem sustainable, not punishing the everyday user. Small payments & small merchants stay free, and only large merchant transactions carry a nominal charge — still cheaper than global card networks. #UPI #FinTech #DigitalPayments #NPCI #IndiaFinTech #UPIMDR #ZeroMDR

11

3 Comments

What if you could bring GPU-level LLM inference to CPUs? I was recently invited to Kompact.ai’s global launch event in Bengaluru. Great to see such ambitious work coming out of India’s AI ecosystem. 🚀 Kompact AI by Ziroh Labs is building scalable LLM inference on CPUs—without compromising model quality, while making high-performance inference more accessible and efficient. GPU power. CPU scale. No compromise. Thank you to the Kompact.ai team for the invitation and for putting together a great launch event. 🚀

653

6 Comments

Today our India endpoint is generally available! Until now, building with us in India meant routing audio offshore or running everything self-hosted yourself. api.in.deepgram.com runs in Hyderabad. Storage and inference for your audio, transcripts, and speech output happen in India. Same API, same models, same SDKs as our Global, EU, and Australia endpoints. Pricing matches global rates. Moving over is a base URL change. Your keys and integrations keep working. No waitlist, no enterprise-only gate, available to every customer today. Blog post and links in the comments.

53

2 Comments

🚀 Downtime hua hai? Parallel setup. 🔐 Security breach hua? Parallel setup. ⚙️ Performance testing hai? Parallel setup. 📈 Optimization karna hai? Parallel setup. 🤯 Kuch samajh nahi aa raha? …well, you already know the answer. 😄 Some days it feels like parallel setup is the official “Ctrl+Z” of engineering - the magical solution we wish came built-in with every platform and team. But beneath the sarcasm is a simple truth: Reliability isn’t built by wishes… it’s built by good engineers and even better architecture keeping resilience and security on top. And as they say: “Lakh dukhon ki ek dawa hai kyun na aajmae, Parallel setup banae, issue door bhagae.” If only production issues ran away as fast as this poem promises. But until that day arrives… We’ll keep designing redundancy, building resilience, and smiling through incidents like true warriors. 😄 Key principles to achieve this include: - Redundancy leads to Reliability - Automation fosters Confidence - Observability brings Clarity - Cultivating a strong Culture builds Resilience and Security. Successful systems are crafted by forward-thinking teams not the ones sticking with the history. Cheers to engineering teams transforming challenges into solutions and solutions into robust platforms. Let’s persist in enhancing reliability, scalability, security, and trust through intelligent design choices. #EngineeringLeadership #DevOps #SRE #Reliability #Scalability #TeamCulture #TechHumor

17

🇮🇳 𝗠𝗲𝗶𝘁𝗬’𝘀 𝗗𝗲𝗲𝗽𝗳𝗮𝗸𝗲𝘀 𝗰𝗼𝗺𝗯𝗮𝘁: 𝗦𝘁𝗿𝗮𝘁𝗲𝗴𝗶𝗰 & 𝗣𝗹𝗮𝘂𝘀𝗶𝗯𝗹𝗲 𝗕𝗹𝘂𝗲𝗽𝗿𝗶𝗻𝘁 𝗳𝗼𝗿 '𝗨𝗻-𝘃𝗮𝗻𝗶𝘀𝗵𝗮𝗯𝗹𝗲 𝗠𝗲𝘁𝗮𝗱𝗮𝘁𝗮' MeitY’s call for “un-vanishable metadata” opens a new chapter in India’s digital authenticity journey. The task ahead is to build a practical, scalable, and policy-aligned architecture for permanent synthetic traceability. 𝗦𝘆𝗻𝘁𝗵𝗲𝘁𝗶𝗰-𝗔𝗴𝗲𝗻𝘁𝗶𝗰 𝗥𝗲𝘀𝗶𝗹𝗶𝗲𝗻𝗰𝗲 (SAR) framework offers a 𝗽𝗹𝗮𝘂𝘀𝗶𝗯𝗹𝗲 𝗮𝗻𝗱 𝗽𝗿𝗮𝗰𝘁𝗶𝗰𝗮𝗹 𝗯𝗹𝘂𝗲𝗽𝗿𝗶𝗻𝘁 for this mission, embedding authenticity within synthetic content, enabling trust, traceability, and responsible AI innovation 1️⃣ 𝗦𝘆𝗻𝘁𝗵𝗲𝘁𝗶𝗰 → 𝗩𝗶𝘀𝗶𝗯𝗶𝗹𝗶𝘁𝘆 𝗠𝗲𝗶𝘁𝗬’𝘀 𝗔𝘀𝗸: Synthetic data must carry a clear, permanent identifier that cannot be erased or altered 𝗦𝗔𝗥-𝗥𝗲𝘀𝗽𝗼𝗻𝘀𝗲: The Synthetic layer forms Defensive surface. Every AI-generated image, video, or audio is labelled with its Self-Authenticating Signature (Infrangible QR Code-𝗜𝗤𝗥𝗰). This visible yet subtle identity mark ensures transparency and awareness, aligning with MeitY’s 10 % visibility guideline 2️⃣ 𝗔𝗴𝗲𝗻𝘁𝗶𝗰 → 𝗚𝗼𝘃𝗲𝗿𝗻𝗮𝗻𝗰𝗲 𝗠𝗲𝗶𝘁𝗬’𝘀 𝗔𝘀𝗸: Platforms must demonstrate proactive responsibility and due diligence in managing synthetic content 𝗦𝗔𝗥 𝗥𝗲𝘀𝗽𝗼𝗻𝘀𝗲: 𝗔𝗴𝗲𝗻𝘁𝗶𝗰 𝗹𝗮𝘆𝗲𝗿, driven by 𝗖𝗼𝗻𝘁𝗲𝘅𝘁 𝗘𝗻𝗴𝗶𝗻𝗲𝗲𝗿𝗶𝗻𝗴, provides intelligent governance. It verifies metadata instantly, identifies synthetic content at upload, and supports the compliance workflows platforms need to meet MeitY’s due-diligence expectations 3️⃣ 𝗥𝗲𝘀𝗶𝗹𝗶𝗲𝗻𝗰𝗲 → 𝗣𝗿𝗼𝘁𝗲𝗰𝘁𝗶𝗼𝗻 𝗠𝗲𝗶𝘁𝗬’𝘀 𝗔𝘀𝗸: Build an ecosystem structurally secured against 𝗱𝗲𝗲𝗽𝗳𝗮𝗸𝗲𝘀, 𝗧𝗿𝗮𝗰𝗲𝗮𝗯𝗶𝗹𝗶𝘁𝘆 must remain intact 𝗦𝗔𝗥 𝗥𝗲𝘀𝗽𝗼𝗻𝘀𝗲: Resilience layer ensures continuity and protection. Its unerasable IQRc signature and DRM (Data-Container) preserve each asset’s fingerprint and authenticity trail, providing enduring proof even when content is reshared or republished This integrated approach evolves compliance into a 𝗡𝗮𝘁𝗶𝗼𝗻𝗮𝗹 𝗦𝘆𝗻𝘁𝗵𝗲𝘁𝗶𝗰 𝗔𝘂𝘁𝗵𝗲𝗻𝘁𝗶𝗰𝗶𝘁𝘆 𝗙𝗿𝗮𝗺𝗲𝘄𝗼𝗿𝗸, a plausible, practical model to secure digital truth in the synthetic era At its core, the Context-Engineering Layer preserves the ‘सार’ (𝗦𝗮̄𝗿), the vital truth of original data. Even if artefacts are manipulated, the authentic source remains shielded Together, these layers turn MeitY’s vision into an actionable, nationally scalable framework 🔊 𝗙𝘂𝘁𝘂𝗿𝗲 𝗘𝘅𝘁𝗲𝗻𝘀𝗶𝗼𝗻𝘀: 𝗧𝗵𝗲 𝗦𝗼𝘂𝗻𝗱 & 𝗩𝗶𝘀𝗶𝗼𝗻 𝗼𝗳 𝗧𝗿𝘂𝘀𝘁 Beyond visuals, Perceptual Threshold (PerTH) audio watermarking embeds inaudible neural signatures below the human hearing limit, imperceptible yet extractable for authenticity verification As MeitY seeks feedback till Nov 6, collective action can make this vision real -protecting digital truth and advancing responsible AI #MeitY #DigitalIndia #AI #Deepfake #SyntheticContent #SAR

6

1 Comment