TrendyTech

Everything I do at Trendytech is just driven by just 1 goal 
'Changing Lives of my students with Data Engineering & Gen-AI'

Founder of Trendytech

ValuWiki

I’m supporting ValuWiki on the Data Engineering side - building reliable pipelines that keep research data organized and accessible.
On top of that, I’m advising the team on Generative AI, especially RAG tuning, to make the product smarter and more trustworthy.
Excited to bring my data + AI experience into a space where precision and clarity matter the most.

Technical Advisor

VMware

• Led the design and implementation of VMC on AWS Analytics, enhancing cloud data management capabilities.  
• Developed scalable data engineering pipelines utilizing cloud services, improving data processing efficiency.  
• Established robust health monitoring and alerting systems, ensuring optimal performance and reliability.

Member of Technical Staff, VMware

Cisco

• Developed and designed robust Big Data systems to analyze extensive sales data, driving actionable insights for decision-making.  
• Spearheaded Anaplan data integration initiatives from India, enhancing data flow and operational efficiency.  
• Collaborated with cross-functional teams to ensure seamless integration and alignment with business objectives.

Big Data Developer, Cisco

Hello Data Enthusiasts,

I am Sumit Mittal (Founder @ Trendytech.in), offering In-depth Big Data on Cloud & GenAI Trainings.

An NIT Trichy alumnus, I have worked with Top companies like Cisco & VMware before starting Trendytech.

My passion towards transforming lives of students, made me able to train:
- 30,000+ IT Professionals
- Working at Top Product based Companies like Microsoft, Walmart, Google, VISA, Mastercard and more

My structured way of preparing, ensures to give you exposure equivalent to a 4+ years actually experienced Data Engineering professional.

I have empowered Individuals to Transition & Grow Exponentially, through my below most popular programs:
-  Ultimate Data Engineering Program
-  Gen-AI Masters Program

Entire curriculum is designed basis industry requirement, covering most important tech stack like:
- Pyspark, Databricks, Data Factory, AWS Cloud Services, Data Modeling, System Design & a lot more
- Agentic AI, RAG, LangChain, Prompt Engineering and more

Do Visit my Website https://trendytech.in/ for more details

Sumit Mittal

For a very limited time, my entire 50-Days SQL Superstar Program is completely free.

The earlier SQL playlist helped millions, but interviews today need more depth. 

So I am rebuilding everything from scratch to help you crack top product based companies.

I am also organising it in one clean portal with daily videos, notes, datasets, quizzes and certificates.

18 days are already live.

Enrolment Link: https://lnkd.in/giyJuJeG

Free access available for a limited time.

These 50 days can change your life!

#sql #dataengineering #databases

Founder @ TrendyTech.in | Data Engineering & GenAI Mentor | 400K+ Followers | Shaping the Next Generation of Data Engineers

5 Free Courses that you can't miss if you are an aspiring Data Engineer.

1. Python for Data Engineers - https://lnkd.in/gwsTQRgR

2. Complete SQL For Data Engineers - https://lnkd.in/gEj-vWa9

3. Data Engineering mock Interviews - https://lnkd.in/gRNgf2WM

4. Azure Data Engineer Project - https://lnkd.in/gUeKwATM

5. Git & Git Hub for Data Engineers - https://lnkd.in/g6GTZh2D

I hope this will help all the aspiring Data Engineers.

PS~ My new Data Engineering Batch is starting on coming Saturday. DM to know more!

#bigdata #databricks #apachespark

Switch from Service based to Product based company

Cracking 4 offers and getting desired hike

Transition from ETL to Data Engineering after 6 years

Joining one of Big 4 companies, Deloitte with 100%+ hike

This is a glimpse of what my students shared with me on their achievements last week.

So Proud to see all of them achieving their goals with their hard work, consistent efforts and never giving up attitude.

- M S Naga Chaithanya switched from Wipro and joined product-based company Amgen India as Manager-Data Engineering.

- Apurva Gawande cracked 4 offers and joined EPAM Systems as Data Engineer.

- Anirudh Mathur after spending 6 years in ETL, now joined EPAM as Data Engineer.

- Ranvijay Singh joined Deloitte as Consultant Data Engineer.

Just one message to every Data Enthusiast in the closing month of 2025;
If these and 1000s of my students can transition and take their career years ahead, you too definitely can.

Trust yourself, have faith in almighty and give it a try.
There might me some delay but Success is 100% bond to come one day.

Some success stories don’t just inspire, they remind us why courage matters.

Journey of my student, Sanietahera is one of them.

Working as a Hadoop Developer at TCS, she felt her growth slowing down. Many people accept that reality but she didn’t.

With uncertainty, fear and a strong desire to grow, she made a tough decision -
to rebuild her career in Data Engineering from the ground up.

She strengthened her fundamentals, mastered cloud, became confident in PySpark, optimisation techniques and Azure Databricks, and went on to deliver complex projects at Fractal.

Recently her efforts turned into yet another milestone.

She joined Coforge as Lead Data Engineer with an exceptional hike.

When she says - Sumit Sir’s Big Data course helped me get industry ready and become a solid Data Engineer;
This fills me with lot more satisfaction that my teaching is impacting lives and career of many.

Her story is a reminder that:
- Progress begins the moment you stop holding yourself back
- Skills can change destiny

Proud of your journey, Sanietahera. Wishing you even bigger wins ahead.

For those who are looking to make a career in Data Engineering. My new batch is starting on coming Saturday. DM to know more.

Most people use Spark every day but very few understand what actually happens behind the scenes.

Let me explain it in the simplest way possible.

Lets say you have a 20 node spark cluster

Each node is of size - 16 cpu cores / 64 gb RAM

Let's say each node has 3 executors,
with each executor of size - 5 cpu cores / 21 GB RAM

Now let's answer the 5 below questions.

=> 1. What's the total capacity of cluster?

We will have 20 * 3 = 60 executors
Total CPU capacity: 60 * 5 = 300 cpu Cores
Total Memory capacity: 60 * 21 = 1260 GB RAM

=> 2. How many parallel tasks can run on this cluster?

We have 300 CPU cores, we can run 300 parallel tasks on this cluster.

=> 3. Let's say you requested for 4 executors then how many parallel tasks can run?

so the capacity we got is 20 cpu cores
84 GB RAM
so a total of 20 parallel tasks can run.

=> 4. Let's say we read a csv file of 10.1 GB stored in datalake and have to do some filtering of data, how many tasks will run?

if we create a dataframe out of 10.1 GB file we will get 81 partitions in our dataframe. (will cover in my next post on how many partitions are created)

so we have 81 partitions each of size 128 mb, the last partition will be a bit smaller.

so our job will have 81 total tasks.
but we have 20 cpu cores

lets say each task takes around 10 second to process 128 mb data.
so first 20 tasks run in parallel,
once these 20 tasks are done the other 20 tasks are executed and so on...

so totally 5 cycles, if we think the most ideal scenario.

10 sec + 10 sec + 10 sec + 10 sec + 8 sec

first 4 cycles is to process 80 tasks all of 128 mb,
last 8 sec is to process just one task of around 100 mb, so it takes little lesser but 19 cpu cores were free during this time.

=> 5. is there a possibility of, out of memory error in the above scenario?

Each executor has 5 cpu cores and 21 gb ram.
This 21 gb RAM is divided in various parts -
300 mb reserved memory,
40% user memory to store user defined variables/data. example hashmap
60% spark memory - this is divided 50:50 between storage memory and execution memory.

so basically we are looking at execution memory and it will be around 28% roughly of the total memory allotted.

so consider around 6 GB of 21 GB memory is meant for execution memory.
per cpu core we have (6 GB / 5 cores) = 1.2 GB execution memory.

That means our task can roughly handle around 1.2 GB of data.
however, we are handling 128 mb so we are well under this range.

Once you start seeing Spark this way, everything becomes simpler.
You understand why jobs run slower, 
why skew happens, 
how parallelism works, 
and how memory gets used.

This is the foundation behind writing efficient Spark code.

I hope you liked this scenario based explanation :)

#bigdata #apachespark

One more student joining as Senior Data Engineer!

congratulations Madhumitha S, I am really happy for you.

A message to all who feel they are stuck,

Keep working hard with right guidance & you will be amazed to see the results.

New Milestone Unlocked – Joined Tech Mahindra as Senior Data Engineer with 100% hike

If you’re in IT field, one thing is clear — continuous upskilling is not optional anymore.

A few years back, I was working as a System Engineer with limited exposure.

But I took a bold decision to transition towards Data Engineering — and honestly, that changed everything for me.

From learning Spark, optimisation techniques, complex data pipelines, Databricks & AWS through Sumit Mittal  ’s Big Data Program;
I got the confidence and the skills to land my first big break in this exciting domain.

Were weekends spent studying? Yes.
Was it worth it in the end? Absolutely yes.

Today, after gaining multiple project experiences, I’m super thrilled to share that I’ve joined Tech Mahindra as a Senior Data Engineer with a great hike.

A big thank you to everyone who supported me throughout this journey—and a special thanks to Sumit Sir, whose teaching style kept me motivated every single day.

The growth journey continues and a long way to go further.

The enrolments for my Free 50-Days SQL Superstar Program will close soon.

Over 7000 candidates have enrolled for it.

If you’ve been planning to start, this is the best time.

All video sessions, notes, datasets, quizzes, and certificates are already organised on the learning portal.

Once the window closes, new registrations won’t be accepted.

Enrol here: https://lnkd.in/giyJuJeG

These 50 Days can change your life!

Do Like & Repost so that maximum people can benefit from this.

#sql #database #dataengineering

What things in Gen AI a Data Engineer should know!

LLM - Large Language Model (act's like a brain)

This LLM has 2 main stages
- Pre training
- Fine tuning
although there are more.. 

In Pre training Big Companies will take a big chunk of data (100's of TB) available on internet, blogs, wikipedia etc and get a gist of it in a few GB's. 

Lets consider 70 Billion parameter model, each parameter here will be a float of 2 bytes and will take 140 GB.

This will take 10's of days and 1000's of heavy machines, and 10's or 100's of million dollars to do this and we get a base/foundation model.

This base model cannot answer question, rather if you give it a question it might give more questions.

Now, we can leverage the foundation model and develop something.

As a Data Engineer, 

=> Prompt Engineering is the easiest thing to do, but when creating applications you need to write it in a programatic way. writing better prompts with more context can help us get better results.

=> Retrieval Augmented Generation (RAG) - As a Data Engineer you can index your company documents, here you would require to clean the data and index it in a vector database.

Later when a prompt is given, it will try to do a simarity search and will search in your vector database. whatever results come will be augmented to your prompt and it's like more context is passed as part of the prompt for better results.

So, setting up a vector DB, cleaning documents, and storing in vector DB can be directly in a Data Engineers portfolio.

=> Model Fine Tuning - This is where the world is moving towards. Instead of striving for bigger and bigger base model, it has been seen that a decent size base model with a good fine tuning done for specific tasks can perform better.

So, we will take a foundation mode, and we might have 10's or 100's of thousands of quality questions and answers (high quality training data) on which we fine tune the model.

There can be different fine tuned models specific to each task.

Do not try using a single LLM for everything.

A better way is to use specialised LLM's fine tuned differently to do different stuff's.

A model might be good at summarisation, other might be good at judging which summaries are better.

Also the LLM's can be given access to various tools like google search, libraries to plot a graph, calculator etc in order to achieve a task.

=> Now when you develop complex application, it might happen you use a LLM for summarisation, and the output of this LLM goes to other LLM for judging. 
So there can be a big pipeline, this can be developed using langchain.

So I feel
- A basic understanding of terminologies around Pre training
- Vector Database
- RAG
- Model Fine tuning to some extent
- Lang Chain
- Agentic
are the things which at the minimum a Data Engineer should know.

But there are many more things within this, which I will cover in extreme depth in my "Gen AI for Data Engineers Program"

I hope you learnt something new!

A few months back, I received a message from a working professional.

He wrote:

"I sit quietly during Gen AI discussions at work. Everyone talks about RAG, embeddings, transformers. I don’t understand anything. I feel I am falling behind."

He wasn’t new to Data Engineering.
He had more than a decade of experience.
Great at ETL, cloud, pipelines.

But when Gen AI conversations started in his team,
he felt invisible.

He joined my program.

Week 1:
He struggled with the terms. Even simple concepts like embeddings felt intimidating.

Week 7:
He built his first RAG application.
He could finally connect how chunking, indexing, and retrieval work.

Week 10:
He presented a prototype to his leadership.
Not perfect, but functional.

Week 14:
The same person who once sat silently in meetings,
now leads the internal Gen AI initiative inside his team.

His exact words:
"Earlier Gen AI looked complex. Now I realize it was just unfamiliar."

He didn’t magically gain confidence.

Confidence followed clarity.

Clarity followed action.

Today he is the go-to person at his workplace for Gen AI.

Not because he is the smartest.
Because he started early and stayed consistent.

If you’re sitting quiet in meetings today,
you could be leading them six months from now.

All you need is a starting point.

DM to know more about my Gen AI Program!

We are implementing a Gen AI based solution where we are planning to ease out the work of technical Support team.

currently students raise a zoho support ticket through the learning portal. Our support team answers it. 

This is a perfect use case where Gen AI can help to ease out the load of support team. Also the turnaround time can be shortened.

Let me tell you how we are approaching this.

we are working on creating a chatbot (initially it will be support team facing), 
Later once tested and enhanced this will be facing the user directly.

We are taking the previous tickets (can be a complete thread of Question, answer, followup question, answer....) using Zoho API

The first time we pick lets say all the tickets for last 3 months.

One ticket(better represented in json as its a complete thread) can be thought of as one chunk.

we are planning to use MongoDB Atlas service as it can store Json efficiently along with embeddings.

so all the documents from Zoho are now in MongoDB and provide us vector search capability.

When a user type in the query we are using a LLM to reframe the question properly and also divide a question into relevant sub questions.

then for each generated question we are doing a vector search and fetching top 2 chunks for each question above a certain threshold.

Now we are framing the right prompt for RAG.

The final answer would be well aligned with the previous answers given by our team.

Challenges Faced:-
1. Web Scraping Zoho Data which took many attempts to fetch the full thread
 
2. Data Cleaning from Text, We used several regular expressions but most would cause data loss. 

3. Initially we tried Managing a database (pinecone) for vector storage and text files. Before we finally streamlined into MongoDB Atlas for simplicity and scalability. we wanted to tickets in Json format as its a full thread.

A few more things -

we are trying to make it a configuration driven architecture. If we want to enable chat history there will be just one config flag which can be set.

Also, we will be managing the prompts in Langchain hub, so that these can be versioned and any changes should not involve code changes.

We are using a smaller model as a LLM, as its very cost effective. Also we do not most likely need a powerful LLM as our answer should be coming from the context only.

Further enhancements -

what if the context is not present, then right now we will say "Don't know"

But going forward if context is not there, we will bring a agentic behavior so that it can do a google search and try answering it.

Also we want to add incremental context.

So both RAG + Agentic Behaviour can make it even more effective.

These days there is a very huge push from the management to work on Gen AI POC's.

Now with this post you will have an idea on what you can try building within your organization.

I hope this post should help all the Gen AI aspirants.

PS~ My new Gen AI batch is starting on Saturday. DM to know more!

HemansAI

Data & AI Consulting, Development, AI Leadership Training.

Founder

85% of AI products and projects fail.

The two biggest challenges businesses face with AI:

1. Estimating and demonstrating AI value [VPs, CEOs, product managers, and decision-makers]
2. Lack of talent and skills [AI engineers and data professionals]

Most business leaders fall into the first category.

Even if you’re skilled at your job, your team might lack confidence that the AI solution will deliver the desired outcomes.

It’s frustrating, right? You see others succeeding with AI, but your team your team is stuck planning, fixing, or juggling disconnected tools.

AI is supposed to bring automation, sharper insights, and smarter decisions.

But when you try to implement it, you encounter chaos:

- Lack of clarity on ROI
- Endless troubleshooting instead of progress estimation
- Too many disconnected systems to demonstrate AI value

It’s easy to wonder if AI will ever deliver what it promises for your business and career.

Now, imagine a future where AI doesn’t just work—it works seamlessly in your organization.

Your data isn’t just stored—it powers real-time insights, precise forecasts, and simplified processes.

Picture an AI strategy that works without constant troubleshooting—a setup that scales and brings results.

I’m an AI consultant, here’s how I can help you:

- the AI leadership coaching

It’s about creating an AI path that’s right for your business or your career with real results, not just ideas.

Let’s connect. we’ll talk about AI strategies, products, and career opportunities.
connect@himanshuramchandani.co

Himanshu Ramchandani

Most AI engineers can't explain why their agents fail in production

I reviewed 43 agent systems this quarter

- only 3 had proper error handling
- only 1 had memory that actually worked
- zero had correct guardrails implementation

Here's the production-ready architecture most engineers skip:

PHASE 1 - FOUNDATIONS

- stop calling APIs
- start building cognitive loops

↳ reAct (Thought → Action → Observation)
↳ structured outputs with Pydantic validation
↳ tool execution layers that don't hallucinate
↳ repos - langchain-ai/langgraph

PHASE 2 - COORDINATION

- single agents hit walls fast
- teams scale infinitely

↳ langgraph state machines for memory
↳ multi-agent orchestration patterns
↳ RAG that cites sources correctly
↳ frameworks: /crewAI, microsoft/autogen

PHASE 3 - PRODUCTION REALITY

- this is where 90% fail

↳ guardrails: guardrails-ai/guardrails, NVIDIA/NeMo-Guardrails
↳ human-in-the-loop for critical decisions
↳ state rollback when execution breaks
↳ cost monitoring and model switching

PHASE 4 - THE FRONTIER

- self-improving systems
- autonomous exploration

↳ tree-of-Thought reasoning: princeton-nlp/tree-of-thought-llm
↳ multi-agent debates for validation
↳ systems that learn from failures: OpenBMB/ChatDev

The complete roadmap with 8 production projects → see comments

Engineers shipping Phase 3 systems - $180k-$250k
Engineers stuck in tutorial hell - $80k-$120k

The architecture gap is a $100k salary gap

Which phase are you currently building?

Follow Himanshu Ramchandani for AI engineering that goes beyond tutorials


P.S. - This roadmap took 200+ hours - Now it's free. The implementation is where you prove mastery. Check the roadmap below. 

PPS: Last day for AI Agent Architect Bootcamp Registration, last cohort for this year.

AI Engineering & Leadership • Building an AI product studio and providing Development & Consulting services

AI Agent Architect Roadmap

YOU copying RAG tutorials from YouTube

But 90% fail in production

I spent 4 hours analyzing GitHub's top RAG repositories

Here's what actually works:

1) RAGFlow - 68.7k stars
production-ready engine with deep document parsing

2) RAG_Techniques - 23k stars
30+ advanced techniques with runnable code

3) LangChain/LlamaIndex
Still relevant, but basics only

Bonus one

4) Awesome-RAG
list of Retrieval-Augmented Generation (RAG) applications in GenAI

What separates winners from copypasters:

↳ they chunk semantically, not by character count

→ 150-word fixed chunks = context loss
→ semantic boundaries = 42% better retrieval

↳ they don't just embed and pray

→ add query rewriting
→ implement hybrid search (keyword + vector)
→ use re-ranking (cross-encoders, not just cosine)

↳ they test on YOUR data

→ generic benchmarks lie
→ test retrieval time <200ms
→ measure answer accuracy, not vibes

The brutal truth:

simple RAG = 5-10 likes on your demo
production RAG = actual users who don't complain


Clone the repos, test and break things

Then build what actually scales

Your chatbot shouldn't hallucinate more than your CEO (don't beat me for this 💀 )

Follow Himanshu Ramchandani for daily AI insights that ship.

P.S. Building RAG systems? Start with evaluation metrics first, then code. Join the FREE AI Agents production workshop today at 7:30 PM IST

Those who were asking what to build in AI. Here are 37 AI projects that'll actually get you hired:

1. Build your own sub-4-bit model quantization library
2. Build your own LLM serving layer with dynamic KV cache
3. Build your own intelligent dynamic batching system for inference
4. Build your own high-performance CUDA kernel library for sparse models
5. Build your own distributed checkpoint optimized file system
6. Build your own stateful AI agent framework with reflective memory
7. Build your own next-generation vector database indexer
8. Build your own RAG orchestrator with self-healing retrieval logic
9. Build your own LLM safety and alignment evaluation framework
10. Build your own universal lossless tokenizer converter tool
11. Build your own serverless, real-time feature store
12. Build your own domain-adaptive synthetic data generator
13. Build your own real-time prediction drift detection system
14. Build your own data version control integrated with feature lineage
15. Build your own adversarial attack defense and mitigation toolkit
16. Build your own automated fairness and bias auditing toolkit
17. Build your own GPU-accelerated explainability engine (XAI)
18. Build your own hardware-aware ML performance profiler
19. Build your own modular reinforcement learning environment
20. Build your own universal model format translator with optimization
21. Build your own multimodal vector database supporting text, image, and audio
22. Build your own vision transformer pre-training library optimized for video
23. Build your own 3D object reconstruction pipeline using neural radiance fields
24. Build your own code generation model from natural language commands
25. Build your own differentiable physics engine for robotics simulation
26. Build your own high-fidelity audio generation library using diffusion models
27. Build your own knowledge graph construction engine from LLM output
28. Build your own data leakage detection system for distributed training
29. Build your own secure model inference sandbox with confidential computing
30. Build your own watermarking system for AI-generated content
31. Build your own federated learning orchestrator with differential privacy
32. Build your own zero-trust model access control layer
33. Build your own Model Card automation tool for governance
34. Build your own system for quantifying model uncertainty in predictions
35. Build your own adaptive GPU scheduler based on convergence speed
36. Build your own serverless platform for rapid fine-tuning jobs
37. Build your own model compiler for specialized commodity hardware

PS: Created a repo: Build Your Own AI Everything
Anyone can contribute to it. By building any project.

You are not stuck. You are just comfortable being mediocre AI professional:

quit tutorial hell. stop copying code. read the paper. implement from scratch. understand backprop. actually understand it. maintain a learning log. write what you built today. deploy something broken. fix it in production. read Attention Is All You Need twice. train a model on your own data. your actual data. not MNIST. feel the GPU burn. watch loss curves at 3am. turn off autopilot. code without stackoverflow. think before you import. eat without scrolling. debug without chatgpt first. plan your architecture. draw it on paper. then on excalidraw. then build. organize your experiments. track every hyperparameter. take AI safety seriously. read the original transformer paper. act on ideas fast. find your niche. build in public. write technical blogs. save someone's training run. learn systems design. read research daily. create your own datasets. stay skeptical of benchmarks. refine your prompts. optimize for latency. act with rigor. help junior engineers. be humble. stop using tools you don't understand. follow your curiosity. craft your portfolio. learn MLOps properly. systemize your experiments. write code. write docs. write posts. write papers. iterate on feedback. leave Discord open less. talk to researchers. contribute to open source. visit AI labs. look for unsolved problems. experience production failures. then learn. sit with your confusion. build something real. embody first principles. what do you actually want to build? turn off notifications. experiment. break things. fine-tune models. fall in love with the problem. notice inefficiencies in pipelines. visualize attention maps. talk to domain experts with humility. don't trash other frameworks. be technical. be precise. become an engineer. cherry-pick your tools. discard the hype. failed experiments aren't failures. invite hard problems. accept when simpler works. read Karpathy. be contrarian. choose depth over breadth. do work that compounds. let problems obsess you. lose yourself in implementation. value compute time. experience the full stack.

what did I miss?

Follow me Himanshu Ramchandani for insights on AI that actually compound

2026 AI Engineering predictions

1) most AI agents fail in production because we have no way to see why they fail and engineers spend most time manually checking outputs

2) RAG systems work great for 3 months then quietly get worse as embeddings drift and data goes stale

3) traditional monitoring tools show system healthy while the AI gives wrong answers because we can track uptime but not reasoning quality

4) bootcamp graduates with shipped projects are getting hired over CS graduates with perfect GPAs because portfolio beats pedigree when the field moves this fast

5) companies need people who can debug production AI systems not people who can explain transformer architecture

6) job posts now ask for agent orchestration and context architecture but nobody agrees what these terms mean yet

7) generic chatbots are becoming commodities while real money is in narrow deep solutions like AI for dental clinics or solar installers

8) small teams building with AI leverage are shipping faster than large traditional teams because less coordination overhead

9) fine tuning is coming back because RAG alone works for simple cases but companies wanting competitive advantage are training their own models

10) multi agent systems sound great in demos but coordination is brutal in production as every agent you add multiplies the failure points

11) context design matters more than I expected because how you structure knowledge and retrieve information determines reliability more than which model you use

12) EU AI Act compliance deadline is August 2026 for high risk systems and if you serve EU customers in healthcare finance or hiring this is real

some questions: 

13) will voice interfaces actually replace typing for coding because it seems unlikely

14) one person with AI agents really compete with 50 person teams maybe for simple products but skeptical for complex B2B systems

15) are degrees becoming worthless or are we in a temporary bootcamp hype cycle because probably somewhere in between (edtech 2.0)

16) I thought prompt engineering would be the key skill but turns out production reliability and system design matter way more

17) I underestimated how much time goes into data pipeline work because it is not glamorous but it is 60% of the actual job

18) I overestimated how fast companies would adopt agents because most are still figuring out basic RAG implementations

19) learn to build systems that work when you are not watching them because production AI needs different skills than prototype AI

20) understanding failure modes matters more than understanding architectures so learn what breaks and why

21) get comfortable with uncertainty because best practices do not exist yet and you will figure things out as you go

What do you see differently? Where do you think I am wrong?

Follow Himanshu Ramchandani for honest takes on AI engineering, not just hype

If you want to ship a production LLM app by 2025, you have less than 40 days so here's what you are going to do:

Stop calling OpenAI and praying it works

Your app breaks when their API goes down

You have zero control

Here's what production engineers actually do:

Build fallbacks

↳ openAI down? switch to anthropic
↳ both down? use your self-hosted model
↳ all automatic - users never notice

The math that changes everything:

- commercial API only:
$9,500/month at 50M requests

- hybrid approach (API + self-hosted):
$730/month at 50M requests

That's 92% savings

Companies like Notion and Shopify don't use one LLM

They use smart routing between multiple models

This is part of the Day 35 of the 75 Hard AI Engineering Challenge:

↳ multi-provider fallback system
↳ self-hosted model with vLLM
↳ cost optimization logic
↳ production deployment on AWS

This isn't theory

The difference between a demo and a product? 

reliability when things break

Build something that actually ships.


Follow Himanshu Ramchandani for real AI engineering.


PS: Build a fridge AI agent that will order groceries automatically, join the community and bootcamp waitlist.

You're watching everyone else build AI agents while you're still watching tutorials

You told yourself "next year"

But next year never comes

Right now? AI Agents are your unfair advantage

The difference between engineers who build and engineers who watch:

→ builders have 7 production agents in their portfolio
→ watchers have 47 bookmarked tutorials they never finished

Here's what hurts:

↳ that customer support role paying $140K? going to someone with agent experience
↳ that promotion you wanted? given to the engineer who automated incident management
↳ that freelance client paying $15K/project? hired someone who already built RAG systems

233 people are waiting against 60 seats - [4 days left]

In 4 weeks, you'll have:

↳ Customer support agent (cuts costs 30%)
↳ Financial analysis agent (10x faster research)  
↳ RAG Q&A system (the 2025 skill everyone's hiring for)
↳ Code reviewer (48 hours → 30 seconds)
↳ Sales qualification agent (4x team productivity)
↳ Incident management agent ($500K+ ROI)
↳ Content generation agent (5x output)

Not theory. Production-ready systems you can show employers tomorrow

Starts December 2nd, 7:30 PM IST

The engineers crushing it in 2025 are building their portfolios right now

Will you be one of them?

Secure your seat here: https://lnkd.in/gtHqDG4c

P.S. Free access to bootcamp orientation + 30-day AI Agents guide + recordings on Nov 30, 9 AM IST
Production Ready AI Agent workshop Satuday 9 AM IST FREE

taplio

The Battle

Sumit Mittal

Himanshu Ramchandani

Followers

Posts

Engagements

Likes

Performance ScoreSumit Mittal's LinkedIn Performance

Performance ScoreHimanshu Ramchandani's LinkedIn Performance

Ready to grow your LinkedIn presence?

taplio

The Battle

Sumit Mittal

Himanshu Ramchandani

Followers

Posts

Engagements

Likes

Performance Score.css-goub6s{font-size:var(--chakra-fontSizes-md);font-weight:500;color:var(--chakra-colors-gray-600);display:block;margin-top:var(--chakra-space-1);}Sumit Mittal's LinkedIn Performance

Performance ScoreHimanshu Ramchandani's LinkedIn Performance

Ready to grow your LinkedIn presence?

Performance ScoreSumit Mittal's LinkedIn Performance