← Back to leaderboard
kmeanskaran

Karan🧋

Verifiedactive
@kmeanskaran

mlops engineer | ai/inference | global ai consultant | your fav tech writer https://t.co/C87steJZyh

RosterTech Career & Coding EducationTrading & MarketsHumor & Meme Accounts
15,039 followers · 496 following · Last polled 3h ago
Perf. Score
53.6
Rank #199
Value Score
62.5
#101 of 273 priced
Median views
1.9K
p25 997 · trailing 7d
Followers
15K
-0.1% 7d
ER (impr.)
3.58%
eng ÷ impressions
ER (foll.)
0.42%
eng ÷ followers
Est. CPM
$25.72
QT-rate basis
WoW views
-13.7%
▼ falling

Rates & value · est. from 1.9K median organic views

fairly priced
Quote tweetbasis
$50
$25.72 / 1K views
Post
—
no CPM
Thread
—
no CPM
Retweet
—
no CPM
Value Score 62.5 · rank #101 of 273 priced$943.4 / 1K engagementsrate age unknown (pre-audit import)

Trends

Follower growth
Median views / day
Engagement rate
View distribution · trailing-7d posts

Each blue dot is one organic post. The Performance Score ranks by the median (1.9K), not the mean (1.8K) — so a single viral post doesn't inflate a creator's reach. The p25 floor (997) is the level 75% of posts clear.

Recent posts · ranked by views

PostPostedViewsLikesRepostsRepliesQuotesBookmarksER
Job market in India is completely collapsed today. A friend of mine attending many interviews from Indeed and Naukri. She said, every time I pass the technical round and in the HR round companies are usually asking for money to secure a job. They are asking minimum 5k and they promise that you will get money back in the first salary. What joke is this? There's a whole world like this outside X, where people still learn MERN stack and applying to a job like it's 2019. No skill update, no online presence, no cold DMs. They are totally playing a luck game! Some of my friends got a job as tech support in night shifts and other paying to HR for a job. Demand is too high in Pune because maximum youth are college students or fresher who are looking for a job. Companies on other hand, don't even know what they want. They just dump senior level JD for 2.8LPA. I think to get a job today, you need an online presence. Including X, articles, posting your work on LinkedIn, and most important is building projects on GitHub that too end to end. If you don't have emerging and latest skills, you will never get an opportunity in this field. You need to stand out with the skills where people don't go because it's hard to learn. Up-skilling and online presence is the only way to stay on top with different options for job. And yes, if company is asking a money from you. You better know where you will be working.frozen16d ago55K413242081971.20%
https://t.co/KU5iSJRQRrfrozen21d ago54K42558137432.27%
Don't hate me, but yes, Indian tech companies have lost their charm completely. No AI research labs, no real tech investment, not even startup programs to fund and support Indian startups. Their main job is to keep Indians employed. Sorry, but this is the truth. As AI grows so fast, everyone can code, and even for large systems you don't need 20 engineers, just 5 skilled ones with AI tools. Neither their stock is growing nor does gen want to work there. If you say you work in these companies, people make weird faces today. This is really sad. A small person like me can only share opinions and observations from day-to-day life. Unless India builds AI labs and emerging tech rather than outsourcing, only then can they survive.frozen14d ago41K71949431742.17%
Understanding GPUs will change your AI Engineering journey beyond agents and LLM. Rise of LLMs and deep learning introduced dedicated GPU hardware which serves your models for training and inference. You don't need to start with C++, CUDA, and Maths, but: > Components of a GPU and roles > How a token flows through hardware > Compute vs memory, the two limits > What FLOPS means in practice > Training vs inference hardware needs > What is CUDA Kernels > Calculating GPU memory from model size > Quantization to fit bigger models > Multi-GPU and splitting large models @huggingface has the feature of calculating FLOps and recommending GPUs for your use case. You must explore. In terms of inference engineering, GPU understanding are much helpful your platform performance and handling multiple requests. Read this article for complete breakdown of GPUs for AI Engineers doesn't require any pre-requisite.frozen21d ago33K56495316834.06%
I took interviews, I attended lots of interview in AI/ML positions. Most of the people don't know how to prep with right strategy. So I covered insights from personal experience which will make you in better shape for interivews. > how you can build a AI/ML projects to get hired > how hiring cycle works in AI/ML hiring > importance of system design today > how to create meaningful resume and not just for ATS > MLOps prep > recruiters expectations in AI/ML interviews and much more. There is a lot of noise on internet and specially on X and most articles don't even carry experiences. I thought why not to write a real experience and here it is. Have a read and comment your thoughts.frozen29d ago28K36246905783.59%
If you want to learn complete AI Engineering stack, follow these stages: Stage 1: How LLMs actually work Tokens, context window, embeddings, prefill vs decode, why every extra token costs time and money. Stage 2: Prompts and APIs System prompts, JSON output, tool schemas, retries, prompt versions. Treat prompts like production code. Stage 3: Retrieval (RAG) Chunk docs, search, rerank, keep data fresh, cite sources, rewrite bad queries. Most “hallucinations” are retrieval failures. Stage 4: Backend for GenAI FastAPI, microservices, auth, rate limits, request queues, Redis caching, webhooks, timeouts. The model is not the whole system. Stage 5: Memory Chat memory, long-term memory, summarize vs replay, reuse KV cache, isolate users. Do not store what you should not store. Stage 6: Agents Plan → tools → act → check. Graphs, human approval, timeouts, circuit breakers. Agents fail in loops. Stage 7: Pick and serve the model Small vs big model, Bedrock vs self-host, vLLM / SGLang, batching, KV cache, quantization. Route cheap first, expensive last. Stage 8: Observability Traces, TTFT/ITL, token cost, cache hits, tool errors, CloudWatch, OpenTelemetry. If you cannot see it, you cannot run it. Stage 9: Quality and safety Eval sets, grounded answers, schema checks, prompt injection, PII, filters, audit logs. Measure quality before users or lawyers do. Stage 10: AWS infra EKS or ECS, ECR, S3, Bedrock, ALB, IAM, VPC. Docker, Kubernetes, Terraform, CI/CD, bash scripts. Scale, failover, quotas. Stage 11: Prove you can ship One real RAG + agent app on AWS. FastAPI backend, Terraform + Docker, CI/CD to prod, latency/cost/quality numbers, one incident writeup. Publish an article and repository. Most people stay stuck watching tutorials. I published a series of 3 articles on design your agentic system. Comment what I am missing, helpful for me as well. Part 2: https://t.co/hzmFr0SUqD Part 1: https://t.co/YRN81sGDrh Give it a read.frozen26d ago26K406642846714.45%
we used to remember numpy and pandas classes and methods. https://t.co/JeUCPsNZkEfrozen20d ago26K1.9K76135477.73%
this update is so cool and it can easily replace lots of people from non-tech corporate and BI. but the twist is, people still use microsoft 365 over google workspace. i bet more than 80% of world's corporate runs on microsoft products. and how gemini agreed on this?frozen15d ago16K204661341.53%
"AI will take SWE jobs." Heard this since GPT-3.5 launched 🤣 Real experience: In real-time production with thousands of users and large systems, you need a human brain, experience, and skills not just next-token predictors. After Opus-4.8, coding agents barely improved. Even with max thinking, you still need system-design understanding. Job POV: Data is large and raw. AI agents alone can't ship a final solution. You can iterate with AI, but you still need thinkers and problem solvers to drive it to prod. Freelance POV: Lots of scope. Clients hand you the problem and data; you can make and break the system with AI to find the best solution. Use AI for a straightforward PoC and iterate. Aim for a working solution, not a perfect one. You're judged by the edge cases AI can't see like what breaks in prod, what users will ask. That comes from experience. You can use AI to fix them, but it still needs you. Human-in-the-loop is the best way to ship the best solution. AI will eat people, not jobs. High-level thinkers who solve edge cases will be in very high demand. Be a problem solver with fast delivery. You'll be the hero. Everyone else should find different jobs.frozen1mo ago13K1479290551.82%
Suddenly, inference engineering vanished from your feed but remains the hottest field. In 2027, no-code agent automation will rise, yet serving LLMs behind the scenes is crucial. Inference engineering is not separate. Learn ML, neural networks, and Transformers together with low-latency, high-throughput LLM serving. I will start a project series on building inference engineering at scale with design patterns. Read the article to begin learning inference engineering. Don't miss this skillset to stay relevant long-term.frozen11d ago8.4K30225613077.64%
Guys, if you still don't know how to build production-ready agents then @googlecloud launched Advent of Agents Season 3. 31 days of whole October, you are building secure and scalable production-grade agents. Daily tutorials and 100% free code with GCP deployments and Gemini integrations. Why you should follow? > Every day of hands-on learning production-grade agents > Deploy your agents on GCP > Learn Google ADK and Gemini CLI > Build real proof of work, not toy agents. > At least take inspiration for your projects Check out: https://t.co/s6IinDy4bgfrozen20d ago7.9K12212301914.13%
Today everyone knows about agents, LangChain, and MCPs. Every job post demands AI agents, so this skill is the bare minimum. The field is already saturated, yet people still aren't getting real opportunities. Why? Companies treat AI, ML, and data as the same department, and because of AI they want someone who can work beyond the code. Saturation is another factor. If everyone knows the same thing, it is not in demand or a valuable skill in tech. So dive deeper into niches, like: > Master MLOps and deployment > Learn platforms and their services like Databricks, GCP, and AWS > Understand LLMs and inference engineering > Reinforcement learning > System optimizations Especially if you come from an ML background, you have a good chance and a stable learning curve to adapt to new skills. One niche and your expertise will bring immense opportunities.frozen17d ago7.8K210171002005.63%
I built a complete agentic ML team 👀🔥 Multi-agent system which dives into the data and communicate each other to build an end-to-end ML pipeline. Analyst answers about EDA. Human only approve. Top harness features: > Agents scans data and features. Create report > Human approval and feedback on train/test, choosing models, and promoting to prod > ML team of agents code just like engineers and save the src directory for each runs > Caching on prompts and tokens to save ~40% > Go live directly serves the model. There is more, I will share about this soon :)frozen20d ago6.1K145161511254.99%
pov: you live in Pune✨❤️ https://t.co/PBPJWifDDAfrozen27d ago5.0K31271066.53%
no way this meme got real🤣. https://t.co/gCBoV5Bhmffrozen17d ago4.6K3732010.93%
Non-tech people are committing features and timelines to the business that they think can be done in two days. Most engineers experience this now. Product and sales people commit features with short timelines and assume AI can do it in minutes, like generating an image with ChatGPT. Development is moving faster with AI, but pushing anything new in a short time is unhealthy. Especially when you are working on agents. With multi-agent orchestration, prompting, and retrieval, you still do not know system behavior until you test many times with different queries. From recent experience, retrieval can fetch the right resources, but bipolar answers, mismatched figures, and similar issues still appear and need back-and-forth testing. Even with Astra and Fable together, you still spend a lot of time experimenting. Changes for one query type must not break others. Business people commit features and expect top-quality results within days. They should know we have limits around sessions, tokens, cloud auth, and repeated testing for confirmation. How do you manage this? Please share your thoughts.frozen28d ago3.5K420100141.89%
javascript is the best investment you can do as ai/ml engineer. I am working on agentic codebase which is mixture of python (where agentic logic is written) and typescript (infra and app service code is written).frozen4d ago3.4K75270223.11%
how to find the real engineers at your workplace: - posting work in public - writing articles - has an audience - focus on learning everywhere - cool and only cares about skills - never chase employee of the month award - share their work with you - never see senior/junior in person if you have such people, you're blesses. he/she maybe your next co-founder, contributor, etc. also you will observe they spend lots of time in learning latest things rather than just office work. always energetic to show you something new.frozen24d ago3.0K98811374.78%
It's simple, guys. Follow engineers who share work like creators, not creators who act like engineers. Very few people actually build and share their work. Connect with people who share projects and learn from real work. Many people don't even know the technology but post for engagement, not contribution. And algorithm don't push such work, don't worry one DM with opportunity will change your life's script. People do watch your work and connect with you for life changing opportunities.frozen17d ago2.9K85620264.14%
Attention is all you need was not just a paper but a true revolution. https://t.co/pXiWK7MXNTfrozen22d ago2.9K102430244.63%
Day 2 of building a complete ML team of AI Agents I designed a harness with minimal and important human in the loop decisions. Agents look into the data, code by themselves, version it and fine-tune models on your decision. You just need to select the best model and promote to production. Agents share the important insights which is mixture of both business and ML with these workflow: Understands the data -> Feature Engineering -> Choose Models -> Train and Test -> Go live The most important part is nothing is hardcoded, agents decide the coding and workflow by observation and target. Re-iterate on the solution until final decision of production. I will be deploying on prod with CI/CD, Docker, and Terraform. Also sharing the entire project and github repo. Keep Learning :)frozen19d ago2.7K62681444.46%
me and claude pushing last moment changes to prod https://t.co/3VGguMFgjRfrozen21d ago2.5K6634083.20%
for the next interviews, you don't need to start prep with arrays again. https://t.co/zP0Q2Se3jtfrozen10d ago2.2K65150173.96%
AI is making our lives easier, but software engineers are still burning out. Why? Like me, many engineers burn out even with AI. It makes coding easy and hardly anyone writes from scratch yet expectations have risen. People assume AI lets anyone build anything quickly, but it demands more experimentation and edge-case testing. Vibe coding requires monitoring logs and tool messages closely, or fixes escalate without control. You get hired for AI-driven fast delivery, so failing questions your talent and triggers pressure. Systems stay up 8-10 hours under surveillance where one AI mistake can corrupt files, logic, or structure. You spend significant time watching and instructing AI back and forth. What do you think? Are you burning out even with AI?frozen6d ago2.2K33260122.41%
Love this discussion @philipkiely We really need more such pure engineering discussions on X.2d ago2.1K51300283.86%
how IT people see each other. https://t.co/F9QdoGCBoYfrozen16d ago2.1K2701071.68%
Today, tech people can understand both the tech and the business requirements from the client side. In future, we will see engineers handling client, product release, marketing and sales. And this will affect non-tech background product managers/leaders.2d ago1.9K8916125.09%
If your friends don't have different political opinions, then you're with fake people.frozen12d ago1.9K5124023.13%
I am building an harness for ML team of agents for fun. What it does: > Pick any data, harness shapes agents and metrics > Core pipeline: Data analysis, feature engineering, model training + fine-tuning, evaluation, serving model > It has skeptic who reviews the output of agents > Asks human the important decisions like: choose the recommended feature processing and model seleciton > Agents does coding in background just like engineers > Logs are visible from agents who are coding BTS > Feature store, model versioning, and saving past runs > You can promote the model from all the runs I optimise the agents and model calls along and built deployment pipeline using CI/CD +Terraform + Docker (my fav setup) Soon sharing the project details and repo so you can try by yourself.frozen13d ago1.8K58241295.15%
AI can write code, scripts, and debug, but deployment is still messy and confusing. I'm planning an article on deploying AI/ML systems to production using AI. Drop your production queries. I'll cover them.frozen8d ago1.8K50350214.37%
That's true. New expectations: using AI, one must know the complete SDLC, project management, QA, deployment, product knowledge, documentation, and direct client discussion. Not a master of all, but molded using AI.frozen11d ago1.7K62110225.04%
The Indian government favoured Starlink, with tie-ups announced with Jio and Airtel a few months back. After the BRICKS summit 2026, the Starlink rejection reversed, while Jio builds its own low Earth orbit satellite. If true, Jio must deliver. As an Indian, I am glad this will create more opportunities in India. This step might open a new phase of self-development and owning such systems, as Russia, China, and the US have.frozen14d ago1.7K3313022.31%
Not even 10% of US households use Starlink. Why would the Ambanis allow anyone into their established monopoly? The US blocks BYD and Huawei, and now opposes Chinese LLMs on national security grounds. X, Cursor, and Grok already have many Indian users with no opposition, but Indians already have their own options and Jio is investing in its own satellite. Why is he desperately escalating this while India is on high alert?frozen13d ago1.6K3408022.69%
Reality check: You will succeed at gambling, but never by taking private institute courses. https://t.co/fLelre53Xhfrozen15d ago1.6K3110042.23%
2026 is a first year of complete vibe coding. Pick any tool or model experience is same but Codex/Claude has bit maturity and detailing. I worked on multiple projects including production but now it saturated and on the other hand no one is coding without AI. any open source projects you will see Claude and OpenAI contributors everywhere. But it still fails, how? 1. you prompt what you see and AI fixes what it reads what about human observation, UX, and blackbox architecture? This comes from lots of experience and solving edge cases. Human intelligence play major roles. 2. You must create documentation in markdown and discussion with AI before coding but point 1 matters here as well. 3. Doesn't matter how great the model benchmarks have, you still go and check results by yourself. Which is a good habit before moving to production 4. AI will require your experience while solving edge cases, 70% time of your project needs this I'll be sharing one article on how to solve product level issues using AI like pro.frozen23d ago1.6K3704093.10%
If you are building a French language agent, would you use English or French for instruction prompts? Assume you are using a frontier LLM.frozen15d ago1.6K1904041.73%
Agree. You don't even need to prompt back and forth. I created one markdown file that works with coding agents and does everything: > building agents > scanning S3 data > debugging > versioning eval metrics > fixing edge cases > just ask me yes or no before pushing code to git and running commands. Today models compact context windows themselves and are faster and smarter than the average SWE.frozen10d ago1.5K31030133.08%
if she genuinely loves you she will pay your claude subscription.frozen19d ago1.5K1805001.58%
Sunday morning, low AQI, and Sinhagad Fort❤️ That's what every Punekar needs to end the long weekend. https://t.co/u13g0I5a8cfrozen18d ago1.3K4313044.02%
A fresh morning morning at Sangameshvar Mandir, Saswaad🤍 https://t.co/RwMVZomW1Afrozen21d ago1.2K3912013.54%
@kmeanskaran cluster just reached 15,000 data points 🎉🎉 Expect more consistent posts on AI, ML, MLOps, and the AI ecosystem. Planning video content explaining AI systems in the simplest terms. Thanks❤️ https://t.co/vWb8OsSvGCfrozen12d ago1.2K48015015.33%
OpenAI after seeing Grok Bots launch: https://t.co/PgfJYssiEqfrozen23d ago1.2K2002011.93%
Next update of Tesla: https://t.co/P9gDOxF97Zfrozen6d ago1.2K3802013.53%
we automate astrology 😭😭 https://t.co/Nk7j3S9lcrfrozen14d ago1.1K2420032.64%
Wishing you and your family a very Happy Shardiya Navratri! 🌸 https://t.co/VgUijpGVnffrozen12d ago1.1K4401014.31%
outside localhost, I shoot and edit with my boys😁❤️ https://t.co/OTtgjxvzVlfrozen18d ago9672902013.31%
Hot take: the next big jump in AI won't come from a bigger LLM. @skyfallai just shared their vision for what comes after the LLM era. What LLMs got right? Next-token prediction at scale gave us fluent language, broad reasoning, and strong task transfer. It works especially well for coding, where outputs are bounded and testable. Where they hit a wall? Running a business is long-horizon, partially observable, and constantly changing. Here LLMs struggle: > Errors compound across long plans
> No way to test "what if" before acting
> Fixed weights go stale as the business evolves Introductioning Enterprise World Models. Skyfall's model predicts how the business changes after an action, not the next token. Department models coordinate through a shared view of the company under an "AI CEO," plans are simulated in a sandbox before going live, and the model keeps learning as things change. LLMs become one component, not the whole brain. The goal is tobfree people from operational grind so they can focus on what matters. Read a full article👇frozen17d ago9601520062.40%
nusrat fateh ali khan is more addicted than any drug, just wait for a clock to hit 12 in midnightfrozen9d ago9151200011.42%
Pune's most beautiful devi temple❤️🪷 Mahalaxmi, Sarasbaug https://t.co/5TOoyopxHkfrozen4d ago8313911015.05%
Virat Kohli is on LinkedIn! Will he post, "I'm excited to announce that I joined the 100 century club along with my inspiration Sachin Tendulkar"? https://t.co/sAru9Q0s6tfrozen9d ago7961802012.64%
next video will be memorable for us 🙃 https://t.co/1Phyzo37tx1d ago7082000002.82%
Oxyrich (Punit Balan) banners in Pune during the Ganesh festival.frozen16d ago677301000.59%