AI/ML News & Innovations Hub

See your serverless: introducing the Headlamp plugin for Knative

Headlamp is an open-source, extensible Kubernetes SIG UI project designed to let you explore, manage, and debug cluster resources. Knative brings serverless workloads to Kubernetes, handling traffic routing, autoscaling, and revision management so teams can deploy and iterate without fighting infrastructure. But operating Knative workloads day-to-day can be difficult, there's still a lot of jumping between the kn CLI, kubectl , and the Kubernetes UI to get a full picture of what's running. We built the Headlamp Knative plugin to bridge that very gap, allowing operators to inspect, understand and act on their workloads all from a single place. This plugin was built as part of the LFX mentorship. Here's a tour of what we shipped. Here is a short walkthrough of the Knative plugin for Headlamp: Integrating Knative resources with Headlamp's map view Headlamp's resource mapping works for Knative CRDs too. You can see how KServices, Revisions, and DomainMappings relate to each other in a single graph view. KService management: edit traffic splits, restart pods, and view logs A KService is the top-level resource in Knative: it manages the lifecycle of Routes, Configurations, Revisions, and everything needed to run and expose your application. The plugin gives KServices a full detail view with an Edit Mode toggle for making live changes to traffic splits, autoscaling annotations, and more. Common actions like viewing the YAML, opening logs, triggering a redeploy, or restarting backin…

The Hot Path Belongs to GBDTs, Agents Own the Cold Path: A Payment-Fraud Benchmark

A reproducible benchmark on latency, cost, and reproducibility, and where agents actually earn their keep. The post The Hot Path Belongs to GBDTs, Agents Own the Cold Path: A Payment-Fraud Benchmark appeared first on Towards Data Science .

Retrofit, don’t rebuild: Agentic overlays for transforming legacy enterprise services

In this technical collaboration between AWS and the authors, we present a pragmatic solution: agentic overlays. Agentic overlays are thin wrapper layers that transform traditional REST-based services into agents capable of participating in A2A interactions. They also expose REST APIs as tools compatible with the Model Context Protocol (MCP). Together, they let enterprises add A2A capabilities to existing REST services without rewriting business logic, without duplicating code, and without running parallel infrastructures. This reduces agent sprawl in the infrastructure by reusing existing services as agents. We provide reference architectures and sample code that show how to build agentic overlays.

US is ‘superhero’, China ‘supervillain’ in global AI contest, American officials warn

US House Foreign Affairs Committee Chairman Brian Mast warned that “America is the superhero” and China the “supervillain” in the contest for global artificial intelligence (AI) leadership on Thursday, just two days after US Treasury Secretary Scott Bessent said America’s “biggest risk” on AI is China getting ahead. The United States and China remain locked in an increasingly competitive race for worldwide AI supremacy, with many American officials concerned that China is eroding the US’ early...

Why Does a Bank Need a Chief Scientist?

This article is brought to you by Capital One . After five years leading natural language understanding and eventually the entire Alexa AI organization at Amazon, Prem Natarajan made a nontraditional move: He became Chief Scientist at a bank. Not just any bank: Capital One, a financial institution serving over 100 million customers, helping everyday Americans manage their financial lives. For Natarajan, a veteran of DARPA-funded research and academia who had watched machine learning evolve from task-specific applications to foundation models, the logic was clear. Some of the most interesting advances in AI research and deployment were shifting from big tech’s horizontal platforms to industry verticals like finance, where the most complex problems aren’t just building models but making AI work under the constraints of real-world customer problems, contextual business knowledge, continuous learning, with an incredibly high bar for accuracy and privacy. That’s also what made Capital One the right place to do it. For decades, the company has been recognized as one of the most data- and analytics-driven financial institutions in the industry. Its business model from the very beginning was built around using data and technology to personalize financial products for customers. A decade ago, Capital One went all in on the cloud and rebuilt its data ecosystem, creating a unified environment for data, compute, and AI and machine learning experimentation. Today, its modern infrastructu…

10 Years of MongoDB Atlas: Built for What’s Next

Nearly a decade ago, I joined MongoDB as a Senior Product Manager to help build the company’s new cloud product, MongoDB Atlas. Our customers had been telling us they wanted to bring MongoDB’s familiar developer experience to the cloud, with the reliability and confidence teams needed to run in production. Atlas was our answer. Today, we’re celebrating 10 years of MongoDB Atlas, the generational data platform for AI applications, and the customers who pushed us to build it. Atlas was shaped in close conversation with those customers and scaled alongside them every step of the way. Today, more than 250,000 builders get started on Atlas every month. Atlas serves more than three trillion queries a day (a roughly threefold increase just since 2023!), and represents 75% of MongoDB’s revenue. Those numbers reflect something more important than growth: the trust builders and customers have placed in us to scale their businesses. That trust was earned by listening closely. Every major capability and architectural investment in Atlas was rooted in what customers asked for: the flexibility and speed of MongoDB’s document model, delivered in a platform that removed operational overhead and could scale with their applications. Over time, Atlas expanded beyond a managed database into a broader data platform, because builders kept asking for more flexibility, more simplicity, and more room to build. That matters even more in the AI era. AI applications create new demands, but the underlyi…

Chanel Hong

Chanel Hong robyn.cherinka… Thu, 06/25/2026 - 12:26 Image Director, Head of People Chanel Hong Chanel Hong is Director, Head of People at Toyota Research Institute (TRI), where she leads talent acquisition, people strategy and operations, employee experience, diversity, equity and inclusion, and learning and development. She focuses on building an inclusive organization that enables impactful research and innovation. Her work includes strengthening TRI’s culture, aligning leadership, and developing systems that enable effective operations, engagement, and growth. Since joining TRI in 2016 as an early employee, Chanel has played a foundational role in shaping the institute’s evolution. As chief of staff to TRI CEO and TMC Chief Scientist Dr. Gill Pratt, she led company-wide planning, drove global initiatives across TRI and Toyota Motor Corporation, and established TRI’s stakeholder relations function to strengthen trust and alignment with key stakeholders. She brings more than 25 years of experience in executive advisory, administration, operations, and corporate communications in the technology sector. Chanel holds a bachelor of arts in art history from Mills College and a SHRM Senior Certified Professional (SHRM-SCP) designation.

datasette-export-database 0.3a2

Release: datasette-export-database 0.3a2 An embarrassingly tiny release. The pyproject.toml had pinned to datasette==1.0a27 , inadvertently making this plugin incompatible with all other Datasette versions. It's now datasette>=1.0a27 instead. Tags: datasette

What if plants could talk?

Now our plants won't shut up. Give your plants a voice: https://github.com/openai/planttalk

ATRC school STEM drive reaches 5,300 students

Students submit over 1,100 research projects in year-long STEM scheme

Brent Hoberman on UK tech post-Starmer: 'Are we going to tax everyone to high hell?'

Can home batteries help save the climate and save you money?

Growing numbers of homeowners are installing batteries that store electricity when it is cheap, which helps balance the grid and cuts emissions, and cheaper plug-in batteries will soon let more people do the same

Optimize model training on Amazon SageMaker AI with NVIDIA Blackwell

This post shows you how to configure training jobs on Amazon SageMaker AI to get the most out of Blackwell’s architecture on AWS. You learn how to select batch sizes and sequence lengths that take advantage of Blackwell’s expanded memory, choose the right precision format for your model size (1B to 64B parameters), and apply activation checkpointing strategically. By the end, you have a practical framework for tuning your training configuration and launching distributed training jobs on P6-B200 instances.

Implementing super resolution by deploying SeedVR2 on Amazon SageMaker AI

In this post, we demonstrate how to implement video upscaling using SeedVR2 on SageMaker AI. We cover the solution architecture, walk through the deployment steps, and show performance comparisons that highlight the quality improvements and processing efficiency you can achieve. By the end of this post, you’ll have the practical knowledge needed to implement this super resolution solution.

Build self-service AWS Health analytics to find actionable health insights with AI agents powered by Amazon Bedrock

In this post, we show you how to build Chaplin (Customer Health and Planned Lifecycle Intelligence Nexus), an open source solution that uses AI agents exposed through the Model Context Protocol (MCP) to provide self-service health event analytics.

Building agentic AI applications with a modern data mesh strategy on AWS

This post shows how to build a governed, serverless data mesh on AWS that provides the secure, scalable data foundation production agentic AI requires.

Exclusive: Pocket FM reshuffle continues as India Country Head Suyog Gothi exits

Pocket FM's India Country Head Suyog Gothi has stepped down from the company after a two and a half year stint. He is the latest senior executive to leave the company over the past month, multiple sources told Entrackr . Gothi joined Pocket FM in December 2023. According to sources, Nitin Verma has taken over Gothi's responsibilities as part of an internal leadership reshuffle. "The transition has already taken place and Nitin is handling the role," said a person aware of the development. "The company is realigning teams and more senior level exits are expected over the next few weeks.” Confirming the development to Entrackr , a Pocket FM spokesperson said, “Suyog Gothi moved on from Pocket FM over seven months ago to pursue new opportunities. This was unrelated to Pocket TV.” However, Gothi’s LinkedIn profile reflects that he served as India Country Head until May 2026. Before joining Pocket FM, Gothi spent nearly seven years at PhonePe, where he led the company's UPI payments business before taking over as Head of Merchant Lending. Earlier in his career, he worked as a Business Analyst at Flipkart between June 2014 and July 2015. His next move could not be ascertained. Queries sent to Gothi did not elicit a response until the publication of this story. Responding to queries on more senior level exits, the spokesperson said, “As with any growing organization, leadership transitions happen from time to time for a range of personal and professional reasons. We are not aware o…

Agentic AI security steals the spotlight at Confidential Computing Summit

For a decade, confidential computing has been chipping away at one of security’s hardest problems: data is well encrypted in transit and at rest, but when a processor works on it, that data sits in memory in the clear, exposed to anyone with privileged host access. “Confidential computing’s aim was to solve this with a trusted execution environment, a subset of the CPU that runs the encrypted workload and handles things like memory encryption,” said Marina Moore , lead security researcher at Edera . For years the field felt like post-quantum cryptography PhD research scientist types agreeing the work is essential, while waiting for it to reach mainstream practitioners. At the Confidential Computing Summit in San Francisco this week, the breakout use case came into focus: agentic AI. Like the web before HTTPS “I was in the really early days of HTTP, and then HTTPS came along pretty quickly,” said Mike Bursell , executive director of the Confidential Computing Consortium . He sees agentic AI where the web sat before certificate authorities and public key infrastructure brokered trust online. “The original agent specifications were not written by security architects,” Bursell said, and “some of it feels in need of refinement.” The gap confidential computing fills is attestation, which provides proof of what runs. The hardware hashes the memory and firmware of a protected execution environment and signs the result inside the chip, Bursell explained, producing a measurement a ver…

Beyond the Straight Line: Choosing Between OLS, Interaction Terms, and Tweedie Regression

Whether you should stick to a classic Ordinary Least Squares regression, introduce interaction terms, or pivot to a Tweedie distribution depends entirely on how your data handles the messy reality of zeros and extreme outliers. The post Beyond the Straight Line: Choosing Between OLS, Interaction Terms, and Tweedie Regression appeared first on Towards Data Science .

UAE investors lead world in AI use, HSBC Finds

98% of UAE investors now use AI tools, but most still want human advice on big calls

Which tokens does a hybrid model predict better?

Smartworks to acquire Singapore-based coworking firm Workstudio Spaces

Smartworks Coworking Spaces has announced the acquisition of Singapore based flexible workspace provider Workstudio Spaces through its wholly owned subsidiary, Smartworks Space Pte. Ltd. The transaction is expected to close in July. Workstudio operates around 26,000 square feet of managed workspace in Singapore. Following the acquisition, Smartworks' footprint in the city state is expected to increase to about 76,000 square feet across four centres, with seating capacity of more than 1,500. According to the company, the acquisition will be funded through resources available with its Singapore subsidiary. Smartworks founder and managing director Neetish Sarda said Singapore remains a strategic market for the company. He added that the acquisition will expand its presence in a high demand micro market and broaden its enterprise customer base. As of March 31, 2026, Smartworks managed around 16.1 million square feet of workspace across 66 centres in 15 cities in India and Singapore. The company primarily serves enterprise clients, including multinational corporations, global capability centres, and large Indian businesses. Smartworks' revenue from operations rose 45% year on year to Rs 520 crore in Q4 FY26 from Rs 358 crore in the same quarter last year. The company reported a profit of Rs 16.6 crore in Q4 FY26 against a loss of Rs 8.3 crore in Q4 FY25, as revenue growth outpaced the increase in expenses.

Real-Time Portfolio Optimization with NVIDIA cuFOLIO

Let’s walk through the NVIDIA cuFOLIO Developer Example. This open source, customizable notebook enables GPU accelerated portfolio optimization by constructing an optimal portfolio from the S&P 500 universe and then backtesting against customizable parameters and portfolios. ➡️ Start now: https://build.nvidia.com/nvidia/quantitative-portfolio-optimization 📥 Code: https://github.com/NVIDIA-AI-Blueprints/quantitative-portfolio-optimization/ 📝 Tech blog: https://developer.nvidia.com/blog/accelerate-large-linear-programming-problems-with-nvidia-cuopt 00:00 Interactive Backtesting Intro 00:11 Quantitative Portfolio Optimization 00:26 Deploy on Cloud (Brev) 00:57 Launchable Setup 01:50 Github 01:57 Run Notebook 02:42 2. CVaR Formulation 03:00 3. Data and Model Setup 04:26 4. Solve CVaR Optimization 07:15 5. Backtest Portfolio 08:09 6. GPU v CPU 09:40 7. Appendix 10:05 Outro #quantfinance #portfoliooptimization #algorithmictrading

Query Your Codebase with DeepSeek V4 and vLLM

We’ve uncovered a master gene that switches on human development

We have identified the gene that, when activated, initiates the developmental programme that results in cells forming a human body

Using Gemini to Create Google Sheets

In this tutorial, we will show you how to use Gemini to create Google Sheets, build a useful table, generate formulas, analyze data, and improve the spreadsheet with follow-up prompts.

Turn Research Papers into Insights with DeepSeek-V4 and SGLang

Understanding the brain with AI-driven explanations and experiments

Researchers introduce generative causal testing, which translates black box models into clear hypotheses and verifies them in the scanner, revealing what specific brain regions respond to in language. The post Understanding the brain with AI-driven explanations and experiments appeared first on Microsoft Research .

The race to understand how and when Thwaites glacier will collapse

The loss of Antarctica’s doomsday glacier would transform our planet. Now scientists are revealing the secrets of this remotest of places, and asking the question: is its demise inevitable?

Why do subgroup TWFE regressions and an interaction TWFE model give different coefficients, when they agree in plain OLS?

I am estimating the effect of a continuous treatment on a continuous outcome using a two-way fixed effects model with individual and year fixed effects ( fixest::feols ) and cluster-robust standard errors. I want to estimate heterogeneous effects by sex, a binary, time-invariant grouping variable. With OLS, I get exactly the behaviour I would expect. m The female marginal effect returned by avg_comparisons() is exactly the same as the coefficient from the female-only regression (-0.00206), and likewise the male marginal effect is exactly the same as the male-only regression (-0.00009). However, the same is not true with fixed effects. I estimate m The fitted coefficients are treatment = 0.00279 treatment:sexFemale = -0.00932 and therefore marginaleffects::avg_comparisons( m, variables = "treatment", by = "sex" ) returns Female = -0.00652 Male = 0.00279 However, estimating separate fixed-effects models gives feols( outcome ~ treatment | id + year, data = subset(dat, sex == "Female"), cluster = ~neighbourhood ) feols( outcome ~ treatment | id + year, data = subset(dat, sex == "Male"), cluster = ~neighbourhood ) with estimates Female = -0.00379 Male = -0.00034 The subgroup samples are additive: Female observations + Male observations = pooled observations. Female individuals + Male individuals = pooled individuals. I also verified that this is not an issue with avg_comparisons() ; without fixed effects in fixest::feols() , the marginal effects and subgroup regressions coincide…

ARX Robotics and Roboneers form ARX Industries to scale unmanned ground vehicle production

German company ARX Robotics and Ukraine's Roboneers today announced the formation of ARX Industries, a joint venture to industrialise the mass production of unmanned ground vehicles. The new entity wi...

The Dev Containers Story: Introducing EelApi for Plugin Authors

Modern development has shifted one old IDE paradigm significantly: Now, not only is it possible that a project is not hosted on the same physical or remote machine as your IDE instance, it could even be that both share the same host but are separated from one another inside isolated environments. If you are a […]

The marketing variable no dashboard can measure

New research suggests your CMO's personal life may shape marketing strategy as much as customer data, with implications for every marketing team. The post The marketing variable no dashboard can measure appeared first on MarTech .

Sofina Ventures offloads Rs 177 Cr worth stake in Mamaearth parent

Sofina Ventures has sold its 1.28% stake in Honasa Consumer, the parent company of Mamaearth through a Rs 177 crore bulk deal on NSE on Thursday. According to exchange data, the Belgium-based investment firm sold 41.78 lakh shares, representing a 1.28% stake in the company, at an average price of Rs 424.07 apiece, valued at Rs 177.2 crore As per the company's March 2026 shareholding data, Sofina Ventures owned a 3.29% stake in Honasa Consumer. Following the sale of 1.28% shares through the bulk deal, its stake is estimated to have reduced to nearly 2%. The deal comes at a time when several venture capital and private equity firms are monetising their investments in listed startups. Recent transactions include Actis’ Rs 371 crore bulk deal in Pine Labs , Alpha Wave’s stake sale in Delhivery , and partial exits by SoftBank and ADIA in Lenskart . In Q4 FY26, Mamaearth reported strong 23% year-on-year growth in its revenue from operations to Rs 657 crore from Rs 534 crore in Q4 FY25, while posting a profit of Rs 69.4 crore during the quarter. Earlier this week, the Gurugram-based company acquired a majority stake in Fluence Pharma , its second acquisition in six months after the purchase of Reginald Men , which marked its entry into the men’s grooming category. Honasa’s shares closed at Rs 421.7 on the NSE on Thursday, valuing the company at a market capitalization of approximately Rs 13,595 crore ($1.51 billion).

How can I estimate the time complexity of training a neural network classifier?

I'm working on a face classifier using YOLO, but for the classification step, we are using a neural network with the following architecture: self.fc = nn.Sequential( nn.Linear(input_dim, 256), nn.ReLU(), nn.Dropout(0.3), nn.Linear(256, 128), nn.ReLU(), nn.Dropout(0.3), nn.Linear(128, num_classes) ) I'm training the network with N classes of 200 embeddings each, which means I have 200*N inputs to the neural network. I want to see if there is way to estimate the time complexity of the training phase of the neural network in function of the number of classes. Thank you!

The Two-Way Energy Conversation: How AI Both Strains and Optimises Data Centre Power

South Africa is rapidly positioning itself as the infrastructure backbone of digital economic growth in sub-Saharan Africa. With the local data centres reaching capacity and many new facilities planned, strategies around artificial intelligence (AI) and cloud are shaping the next chapter in the country’s progress. However, according to DFA, South Africa’s premier enterprise connectivity provider, [...]

Energy For AI, AI For Energy: Designing AI-Ready Data Centres

The data centre industry has evolved through successive waves of innovation, from virtualisation to cloud computing, and now to AI. According to Bloomberg, the market for generative AI is expected to reach USD 1.3 trillion by 2032, while PwC projects that AI could contribute up to USD 15.7 trillion to the global economy by 2030, [...]

What it actually takes to scale AI in Europe: ‘The best founders aren’t building for the next funding round’

EBRD and Microsoft Partner to Help African Startups Adopt AI and Scale

The European Bank for Reconstruction and Development and Microsoft have entered a partnership aimed at helping promising African startups use artificial intelligence to improve their operations and prepare for expansion. The organizations signed a memorandum of understanding during the EBRD’s annual meetings in Amsterdam. The agreement establishes a framework for a pilot programme focused on [...]

The Real Winner of Cursor’s $60B Acquisition Won’t Be AI Coding Assistants

When news broke that SpaceX would acquire Cursor’s parent company, Anysphere, in a reported $60 billion all-stock deal, most of the discussion centered around AI. This was another milestone and enormous valuation, and signal that AI is still bringing enormous disruption. Those reactions aren’t wrong but they can overshadow the bigger story. The real significance […]

AI Bias Is Putting LGBTQIA+ People at Risk

The post AI Bias Is Putting LGBTQIA+ People at Risk appeared first on Partnership on AI .

3 Agents. 3 LLMs. 1 Aging GPU: Engineering Parallel Inference on Bare Metal

Beat the 8GB VRAM limit. Learn how to run three different LLMs on a single 8GB GPU using C++ layer multiplexing and admission control. The post 3 Agents. 3 LLMs. 1 Aging GPU: Engineering Parallel Inference on Bare Metal appeared first on Towards Data Science .

African Mobile Operators Move AI From Strategy to Deployment, GSMA Report Finds

Africa’s mobile operators are deploying AI across a growing range of internal functions — including predictive maintenance, network optimization and AI-driven customer service — according to the Mobile Economy Africa 2026 report published this week by the GSMA. The report cited an Airtel initiative launched in Uganda in April 2025, which the telco described as [...]

Introducing a Recommended Agent in AI Chat, With Codex as the Current Default

JetBrains AI supports multiple coding agents, including Junie, Codex, Claude Agent, and any ACP-compatible agent you bring yourself. Previously, AI users in JetBrains IDEs started in Chat mode and had to choose an agent themselves. As models became more advanced, agents became more capable and their adoption grew. We recognize that agents help users achieve […]

SSH Connections Are Moving to JetBrains Daemon in the Toolbox App 3.6 EAP

Starting with the Toolbox App 3.6 Early Access Program (EAP), SSH connections to remote environments use the JetBrains Daemon (jetbrainsd) by default. We’re making this change available in the EAP first because jetbrainsd is becoming the shared background service for JetBrains tools. Moving SSH there helps us validate the new connection flow before using it […]

Where, when and how to watch the 2026 solar eclipse

This August a total solar eclipse is set to be visible across parts of Europe, while a partial eclipse will sweep across about a quarter of the planet – here’s how to catch it

TokenSpeed-Kernel: Portable APIs and High-Performance Kernels for Multi-Silicon LLM Inference

TL;DR The TokenSpeed-kernel is a standalone, open-source subsystem designed to solve backend complexity in LLM inference. It introduces a clean, layered API and registry system that decouples the high-level runtime...

La métrica que hizo tropezar la estrategia ‘AI-first’ de Duolingo

En abril de 2026, el CEO de Duolingo, Luis von Ahn, reconoció que la compañía había retirado uno de los elementos más delicados de su estrategia de inteligencia artificial: el uso de IA dejaba de contar en las evaluaciones de desempeño de sus empleados . Lo llamativo es que, un año antes, una crisis pública en toda regla no había conseguido cambiar su estrategia ni un milímetro. El primer debate se abrió en la primavera de 2025, cuando Duolingo se declaró ‘AI-first’ . Ahí saltó la discusión habitual de la IA frente a las personas. Prendió rápido: usuarios borrándose la app y las redes de la marca inundadas de críticas. Von Ahn resolvió con oficio la crisis reputacional: aclaraciones, matices y un tono más suave. Le funcionó. El fuego se apagó, la estrategia siguió intacta y la empresa continuó creciendo. Pero se había abierto un segundo debate, menos visible pero igualmente importante: el de la evaluación de los empleados. Ese no se aplacaba con una nota de prensa. Fuera apenas trascendió: lo que una empresa haga con sus evaluaciones internas no provoca bajas masivas ni incendia TikTok. Dentro fue otra cosa. No hubo clamor, pero sí una objeción de fondo. Y esta vez el CEO cedió. La comunicación fue casi inversa a la del año anterior: no hubo gran rectificación pública ni operación de imagen. Von Ahn lo mencionó casi de pasada en un podcast: esa métrica se había retirado. Una crisis pública no movió la estrategia . Una objeción interna, sí. Lo interesante no es tanto la difer…

Repositioning retail for the AI era

Artificial intelligence is rapidly reshaping retail, but not in the ways consumers might immediately notice. The biggest transformation may not be flashy virtual try-ons or chatbot shopping assistants, but in how decisions are made behind the scenes: how products surface in search results, how inventory moves through supply chains, how engineers ship code faster, and…

Pocket FM’s AVP Content Ankit Singh exits amid leadership changes

Ankit Singh, Assistant Vice President of Content at Pocket FM, has announced his departure from the company after a two year stint. In a LinkedIn post, Singh said he moved from working on retention, revenue and analytics to leading Pocket FM’s global content marketing function. He said his team managed content marketing across international markets and adopted generative AI for content production, brand campaigns, the Discover platform launch and other initiatives. "In the next chapter, I'm working on something of my own with a close friend for people in the middle of a job search," said Singh. His exit came on the same day the shutdown of Pocket TV, Pocket FM's microdrama vertical, came to light. Responding to Entrackr's queries, the company said Pocket TV had been launched as a beta experiment and was concluded around eight months ago. It also reiterated its focus on its core audio business and global expansion ahead of a potential IPO. Singh's departure also comes amid a series of senior leadership exits at Pocket FM in recent months. Last month, Chief Financial Officer Anurag Sharma stepped down after nearly three years with the company to pursue entrepreneurial opportunities. During the same month, Senior Vice President Mayank Sancheti also stepped down from his role. Pocket FM has also begun discussions to shift its holding company back to India through a reverse flip as it eyes a public listing in the country. Update at 6:10 PM, June 26 : The story has been updated to…