Since 2011 I've spoken about data, AI, visualization, and education at events & organizations.
This YouTube playlist compiles most public talks.
AI makes pictures cheap to create but harder to trust: the reader may be another AI, and a chart, map or CAD model that looks right still needs an independent check.
TDS treats exams as the curriculum and agents as executors: students learn to specify, orchestrate, collaborate, notice, adapt and verify while the tools keep changing.
Treat AI advice, models and agent workflows as hypotheses: test them on your own work, mine your logs for evidence, and make verification part of the workflow.
AI is too weird and fast-moving to trust by intuition alone: question advice, verify with a second model, calibrate confidence, benchmark what matters, and turn surviving evidence into deterministic rules.
Treat AI as a collaborator, not a vending machine: ask it to interview you before it builds a lesson plan, log what actually happens after you use its output, and benchmark any fix before trusting it.
Build AI products around evidence, not ideas: prototype quickly, test with agents and real users, and iterate until the product proves its value.
As AI takes over production, human value shifts to judgment, context, and verification. The real bottleneck is the cost of checking AI's work.
Coding is becoming specification and review: describe the outcome, let agents build it, and focus on whether the result works.
AI can generate charts and analysis. The enduring skill is knowing what deserves to be visualized, challenged, and communicated.
Tools change faster than curricula. The course is really about learning how to learn, automate work, and judge results rather than memorizing tools.
Managing AI resembles managing experts: specify work clearly, verify outputs systematically, and borrow proven practices from professions that already manage complex judgment.
Agentic tools can perform much of a data scientist's research and analysis cheaply. The leverage comes from framing the question and checking the result.
Teachers should use AI as a participant in lesson planning, explanation, assessment, and reflection—not treat it as a separate subject or shortcut.
Data increasingly serves agents rather than dashboards. Structure it so agents can research, analyze, benchmark, and act reliably on behalf of people.
When AI makes charts abundant, curation—not production—becomes scarce: deciding what deserves attention, what is trustworthy, and what fits the audience.
AI makes sophisticated analytical methods cheap and accessible. The scarce skill becomes choosing the right analysis, questioning assumptions, and judging whether results matter.
AI can design and execute workflows while a session is running. The useful skill is orchestrating tools, context, and verification rather than following a fixed script.
Data stacks should be designed for agents, not dashboards: expose granular data and judgment-rich context so agents can aggregate and reason for each question.
AI can turn raw data into analyses, narratives, visuals, and alternative explanations. The craft shifts to questioning, verification, and editorial judgment.
If AI can pass an exam, the exam may be measuring the wrong thing. Assessments should reward effective AI use, verification, and real-world outcomes.
AI keeps moving the boundary between disciplines and automation. Education should organize around problems and judgment rather than fixed departmental skill boxes.
AI performance depends less on clever prompts than on context. Give models the right goals, examples, constraints, history, and tools so they can act effectively.
Software is shifting from fixed interfaces and workflows to agents that interpret goals and choose actions dynamically. Products should be designed around delegated outcomes.
AI breaks the old link between learning, producing work, and proving competence. Education must redesign practice and assessment around what humans should still learn and demonstrate.
AI makes traditional signals of competence easy to fake. Skills and assessments should focus on valuable questions, validation, accountability, and outcomes AI cannot certify for us.
Everyone now has access to an endlessly patient expert. The opportunity is to use AI as a tutor and thinking partner while building habits for checking its mistakes.
Innovation is frontier exploration, not a pipeline with a known destination. Leaders should fund cheap experiments, learn quickly, and expand what becomes possible.
Journalists can use AI to turn curiosity into rapid experiments and explainers. The machine accelerates research and production; editorial judgment decides what is worth publishing.
AI gives individuals on-demand access to research, analysis, and structured advice once reserved for expert teams. Institutions should redesign work around that new leverage.
When AI can generate endless options, design shifts from making artifacts to choosing constraints, framing problems, and deciding what should exist.
Autonomous agents are useful only when their actions can be trusted. Constrain workflows, verify intermediate outputs, and design explicit checks before granting autonomy.
AI can automate research, outreach, personalization, and analysis in institutional advancement. The largest gains come from redesigning workflows rather than adding a chatbot.
When every student has a powerful AI tutor and solver, conventional closed-book assumptions collapse. Assessments must measure judgment, verification, and application instead.
AI is entering every stage of software development. Teams need to redesign the SDLC around faster generation, stronger automated checks, and human accountability.
Students increasingly learn Python by asking AI, copying examples, and debugging interactively. Teaching should embrace these paths while preserving conceptual understanding and verification.
Engineering education should treat AI as infrastructure across the curriculum, not a standalone elective. Courses, assessment, and faculty workflows all need redesign.
Coding agents let humanities students build software without becoming programmers. The valuable skill is expressing intent clearly and critically evaluating what the machine produces.
Modern AI is capable enough to handle surprisingly broad knowledge work. Treat it like a smart junior colleague: delegate aggressively, then review where it matters.
Agents turn LLMs from answer generators into workers that use tools, execute multi-step tasks, and recover from failures. The challenge is control and verification.
Vibe coding treats code as a disposable build artifact. Specify the outcome, let agents implement it, and spend your effort reviewing behavior rather than typing code.
Everyday digital traces—chats, logs, histories, submissions—contain valuable signals. LLMs make it practical to mine them for patterns, stories, and decisions.
LLMs can automate expensive clinical-data work from protocol checks to anomaly detection and enrollment analysis, provided verification is designed into the workflow.
Vibe analysis lets AI handle code and analytical mechanics while you focus on useful questions, surprising insights, verification, and what action follows.
With coding agents, specifications become the product and code becomes cheap. Work faster by delegating implementation, testing aggressively, and optimizing for review.
LLM psychology treats models as minds to probe: test how prompts, personalities, costs, and failure modes change behavior, then use evidence to choose and manage them.
LLMs can generate, interpret, and even invent visualizations. The analyst's role shifts toward asking better questions, testing novelty, and judging whether a visual is useful.
Vibe coding focuses on outcomes rather than code. Give agents clear requirements and enough context, let them implement, and judge delivered behavior instead of implementation details.
Vibe analysis delegates the analytical process to agents and focuses humans on business outcomes. Explore broadly, verify key claims, and automate successful analyses afterward.
LLMs are becoming cheaper, smarter, longer-context, more autonomous, and more multimodal at once. Each trend makes previously bespoke workflows practical to automate.
AI is becoming cheap, capable infrastructure. Use it heavily, manage it like a large team of smart interns, and build verification into the work.
Vibe analysis uses LLM agents across the full data workflow while humans stay pragmatic and skeptical: seek useful outcomes, challenge them, and publish reproducible work.
AI capability is improving far faster than organizational adoption. Use it now, focus on practical outcomes, and expect culture, trust, and workflow redesign to be harder than technology.
DuckDB offers a simpler, faster analytical default than Pandas for many workloads: SQL, low memory use, remote files, rich functions, and embedded deployment.
AI coding works best when repositories are easy for agents to understand, tasks are well-scoped, tests are fast, and humans review outcomes rather than keystrokes.
Almost every task a data scientist does today can be automated using LLMs. The role of data scientists must change.
Code can be analyzed socially, not just technically: repository behavior, collaboration patterns, and AI-based evaluation reveal how people build software and where process breaks.
LLM command-line workflows can automate everyday Python tasks, then grow into agents that execute code and perform reliable calculations with tools like Pydantic AI.
LLMs can turn messy industrial and IoT data into working analyses live, collapsing the gap between domain question, code, model, and decision.
AI is a foundational shift for managers: analytical and production work gets cheaper, so advantage moves to problem selection, judgment, communication, and organizational change.
LLMs can move from a natural-language question to data, analysis, and a finished visualization. The challenge is specifying intent and validating the resulting chart.
Data visualization can become a dialogue with AI: explore alternative charts, critique them, and iterate conversationally until the design communicates the right idea.
LLM command-line tools make Python automation conversational: inspect files, transform data, call APIs, and compose workflows by describing the result you want.
Modern browsers can run substantial Python locally, making interactive tools easier to distribute: many applications need no server, installation, or backend.
Vibe coding lets people build software by describing what they want and iterating on results, even if they do not understand or inspect the generated code.
AI can generate useful software from plain-language instructions. Start with small outcomes, iterate on failures, and treat generated code as something to test rather than trust.
AI can personalize explanation, practice, and feedback for students and teachers, but effective use depends on critical thinking, ethics, and checking what the system produces.
20 Apr 2025. Anand S: The LLM Psychologist Unveils the Hidden Psychology of AI
18 Apr 2025. Visualizing LLM Hallucinations, Anand Subramanian, FOSSASIA Summit 2025
30 Mar 2025. Visualizing LLM Hallucinations, Anand Subramanian, FOSSASIA Summit 2025
26 Mar 2025. Finding actor look-alikes with multi-modal LLMs - Anand S
14 Mar 2025. #FOSSASIA Summit #AI Track about #ArtificialIntelligence and #MachineLearning
02 Feb 2025. Automate Minecraft with Python - Create a Glowstone Pyramid!
09 Jan 2025. SUTD - Automating Data Visualizations using LLMs
08 Jan 2025. PyCon India 2018 - Cleaning data with Python By Anand S
08 Jan 2025. PyConf Hyderabad 2017 - Keynote Session 1 by S Anand
08 Jan 2025. FOSSASIA 2019 - Visualizing Machine Learning by Anand S
08 Jan 2025. PyCon India 2019 - Maps, Delimitation, and Gerrymandering with Python - Anand S
08 Jan 2025. CIR - Interview with CEO Mr Anand S, CEO, Gramener
07 Jan 2025. PyCon India 2021: Exploring Network Clusters in Python - Anand S
07 Jan 2025. NASSCOM XperienceAI Virtual Summit 2021 - DeepTech Workshop 3
07 Jan 2025. How Far Can I Trust A Language Mode? With Anand S
07 Jan 2025. Impact of LLMs (Chatgpt, LLama2 etc) | NLP Podcast #1
07 Jan 2025. PyCon India 2020 Keynote: Making apps seem faster without optimizing - Anand S
07 Jan 2025. PyCon Indonesia 2020: Building deep learning models on geo spatial data to fight dengue - Anand S
07 Jan 2025. Data Storytelling | The Intern Group: Keynote Speaker Series - Anand S
07 Jan 2025. Data Driven Infographics - DFutura - NMIMS School of Design
07 Jan 2025. IIM Ke Baad E10 IIM Gold Medalist journey to a DATA SCIENTIST: Elections, Education, & Energy Fraud
07 Jan 2025. AI as a Learning Tool - What students teachers and parents need to know
07 Jan 2025. LogicLooM 3.0: The Algorhythmic Chase, Marghazi 2025, IIT Madras
06 Dec 2024. Transforming Complex Data into Impactful AI Solutions | Anand Subramanian | TEDxMDIGurgaon
25 Sep 2024. Finding Actor Look-alikes with Multi-modal LLMs
17 Sep 2024. AI for Impact Ep 2: LLMs - The Alien Tech Dilemma with Anand S. [Head of Innovation at Straive]
03 Sep 2024. TEDx MDI Gurgaon, The Psychology of LLMs
31 May 2024. Mastering Prompt Engineering: Insights from Data Visualization Expert S Anand
04 Dec 2023. What's stopping gen ai adoption in enterprises? | Gen AI Unplugged with Anand S
22 Nov 2023. What's new & better in Gen AI than other technologies? | Gen AI Unplugged with Anand S
21 Nov 2023. Gen AI Unplugged with Anand S | Adoption in Enterprises
12 Nov 2023. Anand S - CEO Chat - A Fun Q&A Session
12 Nov 2023. S Anand: Impact of LLMs (Chatgpt, LLama2 etc) | NLP Podcast #1
11 Oct 2023. Art of Storytelling with Data | Anand S, CEO @ Gramener | Leading With Data 01
22 Sep 2023. How Far Can I Trust A Language Model? With Anand S
15 Sep 2023. Leading With Data Ep 01 - The Art of Storytelling with Anand S
13 Sep 2023. Why does Anand S. prefers Purpose Driven Learning? | Leading With Data 01
01 Jul 2022. Exploring Network Clusters - Convergence 2022 Keynote at Flipkart
20 May 2022. Employee Journey - In conversation with Anand S
16 Feb 2022. Exploring the Movie Actor Network in Python
13 Feb 2022. Programming Minecraft with WebSockets in Python
02 Feb 2022. Bar Chart Race in PowerPoint - Tutorial
07 Jul 2021. LIVE \_ Career in Data Science - Degree/Diploma in Programming & Data Science from IIT Madras
18 Jun 2021. How to Create Bar Chart Race Without Code | Build With Gramex by Anand S | S01 E02
18 Jun 2021. Building Data-Driven Infographics With Simple Web Component | Webinar by Anand s
04 Jun 2021. Develop Data-Driven Infographics With Simple Web Components | Webinar by Anand S - CE0, Gramener
13 May 2021. Albus Dumbledore and the Bar Chart Race | Build With Gramex by Anand S | S01 E01
22 Jan 2021. Programming Minecraft with WebSockets in JavaScript
23 Jul 2020. Data as Kingmaker - Kotak: Chasing Growth
01 May 2020. CIR Interview with CEO- Mr Anand S, CEO, Gramener
28 Mar 2020. Visualizing Machine Learning | Anand S
06 Mar 2020. Engaging your audience with interactive PowerPoint decks
03 Mar 2020. [WEBINAR] How to make interactive data driven PPTs | S Anand
16 Nov 2019. Maps, Delimitation, and Gerrymandering with Python - Anand S
27 Oct 2019. Generating comics with JavaScript
06 Sep 2019. Improve Personal Productivity EP-02 | Gramener Podcast
30 Aug 2019. Improve Personal Productivity EP-01 | Gramener Podcast
16 Aug 2019. How can you direct your own Data Movie?
05 Jul 2019. How to Identify your Data Science Roadmap?
27 Jun 2019. Webinar on Visualizing Big Data by S Anand, PGP 2001, CEO, Gramener 20170527 0432 21
26 Jun 2019. Mapping India | Strata NY 2019 | Anand S
10 Jun 2019. How to Build Data Science Teams that Deliver Business Value (Data Summit, Boston)
08 May 2019. S. Anand & Michael Allen | Ujima Radio show | Save Species with A.I.
11 Apr 2019. "What is Emotion Analytics?" Anand and Ganes discuss how AI can drive Data stories
27 Mar 2019. Natural Language Generation by Anand S
27 Mar 2019. Visualizing Machine Learning by Anand S
08 Feb 2019. Krupanidhi Group of Institutions | Krupanidhi School of Management - Mr Anand S (CEO Gramener)
26 Dec 2018. Cleaning data with Python By Anand S
02 Nov 2018. Masterclass on Data Visualization
12 Aug 2018. PyData 2018 - Day 2 - Hall 1
22 Jul 2018. #1 India's Best Data Scientist | Anand S
11 Jun 2018. Visualizing Machine Learning Models By Anand S (Gramener) at DataHack Summit 2017
10 Jun 2018. S Anand & Ganes Kesari Visualising Text
21 Mar 2018. Primetime Debate: 21 March 2018
06 Mar 2018. Employee Attrition Analytics at IIM Bangalore
24 Feb 2018. Masterclass on Data Visualization
03 Feb 2018. Geo-demographic Segmentation IIM Bangalore
22 Dec 2017. Visualizing machine learning algorithms in Python
02 Nov 2017. Visualising Big Data- Module 6
02 Oct 2017. ANAND S of Gramener talks about Automating Analysis @#Cypher2017
01 Aug 2017. What explains our marks?: Anand S
13 Apr 2017. Data Analytics & Market Landscape | Data Analytics Conclave | UpGrad
13 Apr 2017. Data Wrangling & Industry Practices | Data Wrangling Workshop | UpGrad
02 Dec 2016. NxtGen Leadership 2016
21 Apr 2016. BIG DATA PUBLIC LECTURE : Dr. Rajeev Rastogi and S Anand
09 Feb 2016. Top 10 Famous Scientists and Their Inventions | Top 10 data Scientists in India 2015
21 Sep 2015. Anand S Workshop on Visualization at CYPHER2015
19 Sep 2015. Anand S Interview at CYPHER2015
15 Jul 2015. Visualizing the world's largest democratic exercise - Strata London 2015
12 Jul 2015. Faster data processing in Python - PyCon SG 2015
11 Jul 2015. Anand S - Visualising networks
06 Apr 2015. Data, Politics and Anomalies | S Anand | TEDxNMIMSBangalore
13 Oct 2014. Faster data processing in Python
09 Sep 2014. [Part - 3] Hive Meetup - The power equation Big Data + Social in Governance
01 Sep 2014. Scaling real-time visualisations for elections 2014
13 Aug 2014. WCNGT 2014 Day 2 Part 3
13 Aug 2014. WCNGT 2014 Day 2 Part 2
13 Aug 2014. WCNGT 2014 Day 2 Part 1
24 Apr 2014. [ODCBLR2014] Visualizing Election Data by Anand S
01 Apr 2014. Visualising Big Data by Anand S
21 Mar 2014. UNICOM India Big Data Week 2014 Part 7
21 Mar 2014. UNICOM India Big Data Week 2014 Part 6
21 Mar 2014. UNICOM India Big Data Week 2014 Part 5
06 Sep 2013. Data Visualization in PowerPoint with Python by Anand S
27 Jul 2013. Scraping the Government with Javascript - Anand S
04 Jul 2013. Python Pandas tutorial by S. Anand
07 Dec 2012. Anand S - Beautiful visualizations with d3.js
01 Oct 2012. 1st october 2012 thinking with data part 1
01 Oct 2012. 1st october 2012 thinking with data part 3
01 Oct 2012. 1st october 2012 thinking with data part 2
01 Oct 2012. 1st october 2012 thinking with data part 4
17 Sep 2012. Thinking with Data September 17th 2012 Anand part 3
17 Sep 2012. Thinking with Data September 17th 2012 Anand part 4
17 Sep 2012. Thinking with Data September 17th 2012 Anand part 1
17 Sep 2012. Thinking with Data September 17th 2012 Anand part 2
04 Sep 2012. Handling Data (Part 2) - Anand S.
03 Sep 2012. Handling Data (Part 1) - Anand S.
28 Aug 2012. Thinking with Data - Anand S.
27 Jul 2012. The Fifth Elephant - Big data panel
27 Jul 2012. S Anand & Ganes Kesari - Visualising Text
30 Mar 2012. S. Anand: Pictures through numbers
27 Mar 2012. Anand S - Data Visualisation in JavaScript
13 Feb 2012. JSFoo Chennai 2012 Previews - S. Anand - Data Visualisation in JavaScript
If you need a short bio to introduce me, you're welcome to modify this.