About
Website: abhidaas.dev
GitHub: github.com/abhid1234
Blog: abhid.substack.com
X:…
GitHub: github.com/abhid1234
Blog: abhid.substack.com
X:…
Articles by Abhi
-
A quick overview: What's going on with the Fintech boom in India
A quick overview: What's going on with the Fintech boom in India
Banking and financial services in India are on the inception of complete disruption. It is one of the largest unbanked…
50 Comments -
Fierce battlefield in India for Netflix and likes - Part 2
Fierce battlefield in India for Netflix and likes - Part 2
In Part I of this article we have discussed in depth about the market, competitors, and content. In Part II the topics…
6 Comments -
Fierce battlefield in India for Netflix and likes - Part 1
Fierce battlefield in India for Netflix and likes - Part 1
India is a very important space for Netflix and Amazon prime video. As US subscriber growth is tapering off, Netflix…
17 Comments
Activity
11K followers
-
Abhi Das shared thishttps://lnkd.in/gH2zEZNR Model routing is having a moment. OpenRouter just got acquired, and every gateway now promises to pick the right model for every request. I started playing with routers and kept hitting the same wall. Every gateway logs what happened — model, provider, cost, latency. What I couldn't get was the why: which candidates were scored, why the winner won, whether the cheaper model would have held up, and who approved the policy change. So I built AgentRoute — an open-source evidence layer for AI routing. → A typed receipt for every decision: requested, selected, observed, proposed → Counterfactuals are earned by replay under hard budget caps — and when evidence is thin, it abstains instead of guessing → It never touches your traffic and never auto-ships. It drafts the policy change; a human applies it. 👉 Try it in your browser, no accounts: https://lnkd.in/gH2zEZNR 👉 Install: npm install @avee1234/agentroute 👉 Why I built it: https://lnkd.in/gupi-n6D 👉 Code (MIT): https://lnkd.in/g4x-h8i4 If you run OpenRouter, LiteLLM AI Gateway, Cloudflare AI Gateway, or Vercel AI Gateway in production — how do you audit routing today? #AIEngineering #LLMOps #OpenSource #BuildInPublic #DevTools
-
Abhi Das shared thisLooking forward to be one of the superuser of this integration! Congrats team!! Tessa Price Quarles Naina Pasricha Fernanda Chouza Raïssa Tona and many moreAbhi Das shared thisLovable is now available in Google Antigravity. Connect to the platform once and your agents can reach your entire Lovable workspace to build, iterate, query data, and deploy.
-
Abhi Das shared thisA huge thank you to all our partners for such an amazing launch of GE for Industries! After months of hard work and building, we’ve successfully delivered core capabilities across multiple industries. This milestone wouldn't have been possible without the incredible support of our partners and customers, alongside the dedication of our core Product and FDE teams. Looking forward to 100s of customers taking advantage of these industry-specific AI led capabilities that will scale productivity. Thank you Partners - Dun & Bradstreet, FactSet, Moody's Corporation, MSCI Inc., PitchBook, S&P Global, LSEG, Thomson Reuters, Legora , Solve Intelligence, Courtroom5, NetDocuments and many more... Thank you for the amazing partnership: Kristin Thayer Chris Parsonson Zane Harker Sonja Ebron Amy Tsai Shouyi Poon Lisa Darr Azi Handley and many more.. Naveen Punjabi Diane Chang Shikhar Mathur Marc Harpster Steve Sigler Kayla Spiess Hasmig Samurkashian Tomas Moreno Kelly Price Vikram Garlapati Don Tirsell Joshua Gunter Alex Striffler-Hernandez Jake Lehmann and many more... 1) Press Release: https://lnkd.in/gv7DQnfy 2) Thomas Kurian Blog: https://lnkd.in/gHpaRp9p 3) Product Website: https://lnkd.in/gsvr3KFX 4) YouTube Video: https://lnkd.in/gJnFeupT #Google #Gemini #GeminiEnterprise #GoogleCloud #AIAbhi Das shared thisI’m excited to announce the launch of Gemini Enterprise for industries - purpose-built to deliver context, control, and choice for your industry's unique demands. Developed in close collaboration with industry leaders, these offerings combine core capabilities around industry-specific skills, secure connections to trusted systems and data, ready-to-deploy agents, and a robust partner ecosystem. We’re defining the next frontier of business transformation beginning with Financial Services and Legal, with more industries to come. Let’s go! More info: - Gemini Enterprise for Industries overview: https://lnkd.in/g33R8VNw - Gemini Enterprise for Legal blog: https://lnkd.in/gqBZriqu - Gemini Enterprise for Financial Services blog: https://lnkd.in/g-iDRtKP #Google #Gemini #GeminiEnterprise #GoogleCloud #AI
-
Abhi Das shared thisKimi K3 came out and I wanted to actually use it, not benchmark it. A couple of weekends later I had a playground. K3 Planner — give it a goal, "launch a paid AI side project to 100 users in 30 days", and Kimi K3 returns an ordered plan with a risk rating on every step, the assumptions it made, and your first three moves. Free, no signup, running on Fireworks serverless. Demo mode never touches the API, so it works even when the daily cap is spent. kimi-k3-ashy.vercel.app — go break it. Full write-up: https://lnkd.in/g5HuguPZ Code: https://lnkd.in/gX4HHuk8 The model wasn't the hard part. The surfaces around it were. I'd written a "plan strength" score, 0-100: 40 + (steps × 12) + bonuses. Five steps hits the ceiling and every plan has five or more, so it read 100/100 for every plan, forever. Compare mode ranks three strategies by that same score — all three tied, and the "highest score" badge landed on whichever request finished first. A ranking that ranked nothing. A metric that always returns the same number isn't a weak metric. It's decoration. And it's everywhere right now: confidence bars at a fixed width, quality badges computed from nothing. I rebalanced it across five weighted components. The one I'd defend hardest is risk literacy — if a model rates every step "low", it didn't assess risk, it filled a required field. The same sample plan now scores 93. It can disagree with me, which is the only reason to show it. Built on a deliberately small stack: Kimi K3 on Fireworks AI , FastAPI + Pydantic, @Mangum onto Vercel, one hand-written HTML file, 37 tests on GitHub Actions. Worth an audit: whatever confidence indicator your product shows users — what is it computed from?
-
Abhi Das reposted thisAbhi Das reposted thisData teams, RevOps, Finance, Ops, hi 👋 . You already have your data in BigQuery. But the annoying part is that half the company still has to ask someone else to get an answer out of BQ, or they come and ask you. We’re doing a session with Google Cloud on how to fix that. We’ll show how to use Lovable + BigQuery to give teams a simple interface to ask questions, explore data, and get answers without needing to know SQL or file a ticket with the data team. BigQuery stays the source of truth for your teams and projects, and Lovable makes it usable by normal humans. If your company runs on Google Cloud, come hang and learn more! Register here: https://lnkd.in/gaH6hXvD #GoogleCloud #BigQuery #Lovable
-
Abhi Das shared thisThe most interesting part of a making session is usually the part you throw away. You explore a few branches, compare them, pick one, and push it further. At the end, what survives? A transcript, or a single winning file. The alternatives you rejected, and the path between them, are gone — you can't replay them or hand them to anyone. So I built BranchJam: a local, dependency-free browser playground where a session's lasting artifact is an explorable tree of runnable branches, not a transcript. Play a deterministic seed, choose a curated constraint, compare two fixed outcomes side by side, record local votes, then walk the branch tree and inspect any node without disturbing the lineage. What it is not, plainly: outcomes are pre-authored, finite, and deterministic. No live AI, no autonomous generation, no accounts, no network after load — everything stays in one localStorage key on your machine. The proof is a reproducible local run: 35 tests pass, 0 skipped. Invert the core winner comparison on purpose and 3 fail; revert and it's 35/35 again. Why it matters (essay): https://lnkd.in/gbACK2fR Try it in your browser: https://lnkd.in/gw9PX939 Same playground on Hugging Face: https://lnkd.in/gXmDW4-D Code (MIT): https://lnkd.in/gCCf4A6Y The takeaway I keep landing on: branches are worth keeping when they stay runnable. #BuildInPublic #OpenSource #LocalFirst #DevTools
-
Abhi Das shared thisA failing agent trace tells you what happened. It won't tell you which earlier event to investigate. That gap is the whole problem. After a run fails, the trace faithfully records every step — but a final score can't localize the moment things broke, and ordinary inspection blurs the line between an observation and an explanation. So I built Trajectory Ledger: a local, standard-library-only #Python 3.11 CLI. It takes one bounded, inert agent-trace file and produces a reviewable evidence packet. Three ideas do the work: - Evidence-linked lineage: every supported investigation point cites specific ledger record IDs, rebuilt without inventing links. - Explicit abstention: when support is insufficient or ambiguous, it abstains by default instead of guessing a cause. - Fail-closed replay: it compares and reverses one exact saved-task document in a sandboxed process, and fails closed if enforcement can't be proven. The proof is a reproducible local run plus the tests: 52 pass, 0 skips, first result in 0.21 seconds. Break the core recommendation gate on purpose and 3 tests fail; revert and it's 52/52 again. Plainly: this is synthetic-fixtures-only, and makes no causation or efficacy claim. It localizes where the evidence points, or it says so and abstains. Links below 👇 #BuildInPublic #AIAgents #OpenSource #DevTools #AIEngineering
-
Abhi Das shared thisEvery coding agent I use is brilliant and completely amnesiac. It writes code, I correct it, we go back and forth. Then the PR merges and that whole conversation is gone. Next session it starts from zero and I teach it the same things again. Entire (entire.io) keeps it. Every agent session gets stored in your repo, linked to the commit it produced. It's not a GitHub replacement — your repo stays where it is and Entire mirrors it across regions. I got off their waitlist last week. Ran one session on a small library, then asked the file who wrote it: 309 lines: 7% AI (21) · 93% human Twenty-one lines, each traceable to the prompt behind it. So I built grain on top. It reads those sessions and writes your repo's house rules — the test command, the conventions, the things you keep correcting. Open source. Big thanks to Thomas Dohmke and the Entire team for building something genuinely different! Links below!
-
Abhi Das shared thisOur parents' generation is getting robbed — and most tools built to stop it are made for engineers, not for them. So I pointed my agents to build a small thing called Owl 🦉 You paste a suspicious text, email, or call — or just forward it — and Owl tells you, in plain words, whether it looks like a scam, what to do, and what to never do. English or Spanish. Free. No app, no sign-up. It's for anyone, but I really made it for the people scammers hunt hardest: parents, grandparents, and the families who worry about them. Fair warning: I'm not a company, a bank, or a security expert. Owl is a personal project — not financial, legal, or security advice, and not affiliated with any bank or agency. It's a second opinion, not a guarantee, and it can be wrong. When money's on the line, always verify with your bank or someone you trust. And it's fully open source — the whole engine is on GitHub if you want to see how it decides, or build on it: github.com/abhid1234/owl If it helps one person dodge a scam, it was worth building. Links below 👇
Abhi Das commented on a post
1w
Happy Birthday Richard Seroter!
-
Abhi Das liked thisAbhi Das liked thisWhen I work with people on weight loss, one of the most effective strategies I employ is also the simplest: Take a photo of everything you eat. What makes it work long-term is that it is the lowest-friction form of self-monitoring there is. Traditional food logging — weighing portions, entering every ingredient, tallying calories — is accurate on paper but almost impossible to sustain. Most people abandon it within weeks. A photo takes one second. And because it is effortless, people actually keep doing it, which is the only thing that matters. Not to count calories. In fact, forget the calorie number — the apps are often inaccurate, and precision is not the point. The photo is the point. Here is why it works. 1. It interrupts mindless eating. Most overeating is automatic. We eat while distracted, on the go, standing at the counter. Stopping to take a picture inserts a deliberate pause between the impulse and the bite. That pause is where self-control lives. It gives the thinking part of the brain a moment to catch up with the reaching hand. 2. It builds a memory of eating. When we eat without paying attention, the brain fails to properly register the meal — and that weak memory makes us hungrier and more likely to snack later. Attending to your food, even for the second it takes to photograph it, strengthens that memory. Attentive eaters eat less afterward. 3. It is self-monitoring — the most reliable habit in weight loss. Decades of research show that people who consistently track what they eat lose more weight. Here is the surprising part: consistency matters more than completeness. You do not need to log grams and calories. Simply recording that you ate — with a photo — is enough. The people who monitor most often lose the most. 4. It reveals when you are eating off-plan. I also ask patients to set a simple rule: no snacking between meals. The photo becomes an honest mirror. When you reach for something outside your plan, the act of photographing it makes the moment conscious instead of automatic. You catch yourself in real time. The takeaway for anyone trying to lose weight: You do not need a perfect app. You do not need to weigh your food. You need to change your relationship with the moment of eating — from automatic to aware. A photo, taken with attention, does exactly that. Pause. Photograph. That small friction, repeated meal after meal, is where lasting change begins. #WeightLoss #Obesity #BehaviorChange #MindfulEating #LifestyleMedicine
-
Abhi Das liked thisAbhi Das liked thisIt was a pleasure to host and showcase the amazing innovations from Google and many leading cybersecurity ISV partners at Google Cloud's AI Security Innovations Forum this week. Thanks to an incredible set of CISOs, industry leaders and partners, all coming together and discussing key ideas to move the industry forward. More details on the announcements at the event in the blog here - https://lnkd.in/gxiqfka9 Thanks to an amazing set of speakers - Rohan Singla Lionel Litty Siva I. Muhammad Muneer Brice Daniels Josh Madeley Steve Wilson Shree R. Rich Campagna Ramin Farassat Usman Q. Sundar Jayashekar Peter Holko Purvi Desai Joshua Haslett And an incredible set of partners who announced and showcased their new innovations with Google at the event - Acalvio Technologies Britive Check Point Software Cyberhaven Cyera Endor Labs Exabeam Fastly Fortinet Menlo Security Inc. Obsidian Security Palo Alto Networks Ping Identity Qualys Splunk Snyk Thales Cybersecurity Products Transmit Security Zscaler Ashish Verma Anisha Nanda Satish Thomas Kevin Ichhpurani
-
Abhi Das liked thisAbhi Das liked thisI’m super excited to announce that Chronosphere is now Cortex XCOR - our next-generation, AI-native observability platform that provides automated root cause analysis and remediation. Chronosphere was the best placeholder name that Rob Skillington and I could think of on short notice as the company name choice was the main thing blocking investors from wiring us money early on. We thought we would change it later, but we never did - it grew on us and has served us well over the years. It became a huge part of our personal identities and was the title of that rewarding chapter of our lives. It is fitting that we have a new title and name for this next chapter: Cortex XCOR. However, it’s not just a new name / rebrand - it’s the biggest release of our platform ever. One where we reimagined the Observability experience and aimed to deliver the holy grail - a solution that automates root cause analysis and gives you the answer. You’ve probably seen a ton of solutions claiming this along with cool demos, but what impresses me most about what the team has been working on are the results we’re seeing in large production environments: investigations completed in under three minutes on average with a 75% success rate in identifying the root cause. For those who are still concerned about the economics of Observability, don’t worry - we haven’t abandoned our core principles. We’re still helping our customers optimize their data volumes by 89% on average, earning the top ranking Observability platform for cost efficiency and the trust of industry leaders like DoorDash, Affirm and OpenAI. Read more about XCOR in my blog post: https://lnkd.in/gsF7D6cZ
-
Abhi Das liked thisAbhi Das liked thisSay hello to Claude.dev - our new home for developer content on Claude & Claude Code! We are thrilled to launch Claude.dev (https://claude.dev), our new destination for developers building with Claude. Whether you're looking for expert tips on Claude Code or how to be effective with our new models (like Opus 5.5, Sonnet 5.5 and more) this is your new go-to resource. Here's a little of what you'll find on the site: - Engineering deep dives about how we get the most out of Claude for coding - Comprehensive guides for the API and Claude Code (incl upcoming features) - Pro tips and best practices straight from the teams building Claude - A few fun easter eggs hidden along the way! Head over to Claude.dev today and start building something amazing. Let us know what you think! 💻✨ #ai #programming #softwareengineering
-
Abhi Das liked thisAbhi Das liked thisI am thrilled to share the news that I am joining SAS as EVP and Chief Sales Officer. What drew me to SAS is simple: trust. AI is caught in a cycle of rapid acceleration and constant reinvention. As AI moves deeper into how organizations operate, customers need more than access to the next great model. They need confidence in the data, decisions, and outcomes behind it. SAS has built that confidence over five decades by innovating alongside customers and helping them put their most important data to work. I’m excited to help build on that foundation. I am so very grateful for my time at Google and the exceptional colleagues, customers, and partners I had the privilege to work with. I am looking forward to seeing the incredible impact Google will continue to have across the ecosystem.
-
Abhi Das reacted on thisAbhi Das reacted on thisIt’s not a personal post if 2 of 3 people in a photo are colleagues… It’s my first week at JPMorganChase as an Executive Director on the Global Technology Strategy, Innovation and Partnerships team focusing on Cloud & Infrastructure! Thank you Larry Feinsmith and Thomas Frayne for bringing me on. Excited to continue my 15th year in tech business development by fostering cutting edge innovation at the world’s greatest bank. And thank you to Adam J. Epstein, Jamie Hill, Brian QuirkBrian Quirk, and all of my former colleagues at adMarketplace for providing me with a foundational and transformational experience in advertising, especially as you continue to innovate and lead in moving the AI economy forward.
-
Abhi Das liked thisThanks Jesse Zhang and Decagon team - we're excited to meet our joint customers where they are w/ the right data and tools at the right time, and help customers buy seamlessly on our Marketplace. Best is yet to come!Abhi Das liked thisWe’re partnering with Databricks to bring richer customer context into Decagon agents. The best customer experiences depend on understanding the full context behind each customer. But that context often lives across orders, payments, product usage, and other data inside a company’s systems. Through our partnership with Databricks, Decagon agents can access that information directly when they need it, without requiring teams to build and maintain separate data pipelines. Insights from those conversations can also flow back into Databricks, giving companies an even richer view of their customers. We’re also a launch partner for Databricks’ new Transactable Marketplace, making it easier for enterprises to purchase and deploy Decagon through a vendor relationship they already have. Excited to keep building with the Databricks team!
Experience & Education
Licenses & Certifications
Volunteer Experience
Languages
-
Hindi
Full professional proficiency
-
Bengali
Native or bilingual proficiency
Other similar profiles
Explore more posts
Explore top content on LinkedIn
Find curated posts and insights for relevant topics all in one place.
View top contentOthers named Abhi Das in United States
18 others named Abhi Das in United States are on LinkedIn
See others named Abhi Das