Sridhar Seshan

Entrepreneur | Bespoke Solution Consultant | Logistics | BFSI | Supply Chain Consultant

[Sridhar Seshan @sri_s3]

God never gives you a dream that matches your budget. He's not checking your bank account, he's checking your faith.

[Sridhar Seshan @sri_s3]

Your customers face challenges, and your product is designed to address them. However, between the sale and achieving success, something often falters.

This gap-between what you developed and what your customers truly require-is where deals can falter. It's not necessarily due to a flawed product, but rather because deployment, integration, and customization demand expertise from someone who understands both your product and your customer's business.

Forward Deployment Engineers play a crucial role in bridging this gap. They are neither salespeople nor purely support staff; they are engineers integrated into your customer's environment, ensuring the transition from "we bought it" to "we're realizing value from it" is seamless.

Companies that excel in customer success metrics are not simply those with larger support teams; they are the ones that have deployment engineers actively involved, making implementations successful.

This distinction highlights the difference between merely offering a product and fostering a true partnership.

What is the gap between your product promise and your customer's reality?

#CustomerSuccess #ProductDeployment #TechnicalServices #SoftwareImplementation #EngineeringExcellence #ClientPartnerships #B2B #DigitalTransformation

[Sridhar Seshan @sri_s3]

@lets_dig_deeper It's great advice, but same time can you tell me how many of your clients are from colleges and how many companies have benefited using AI. Yes AI has gained importance in today's world

[Sridhar Seshan @sri_s3]

The AI landscape is moving incredibly fast, and we are scaling up to meet the demand.

We are actively expanding our AI service offerings—focusing on helping businesses build intelligent systems, deploy digital labor, and launch AI-driven products rapidly.

If you are looking for a technical partner to help architect your next big idea or need specialized support for your upcoming AI projects, my DMs are open.

Let's build something great together.

🌐 Visit: https://t.co/Um693pOSAr

👨‍💻 Personal Profile: https://t.co/Q6gAgghdnO

#ArtificialIntelligence, #SaaS, #TechInnovation, #AIArchitecture

[Sridhar Seshan @sri_s3]

@Ronald_vanLoon @ultralytics Looks great

[Sridhar Seshan @sri_s3]

#motivation #business #LifeQuotes

[Sridhar Seshan @sri_s3]

After being in hesitation to try @cursor_ai in replacement of @visualstudiocode has provided a better results in my team. People love it but there comes a discussion why this why not @Claudecode. I am looking for your views

[Sridhar Seshan @sri_s3]

Everyone is building RAG pipelines right now.
Most of them will fail in production.

Not because RAG is wrong.

Because they skipped the hard parts.
I've spent the last year building RAG systems on top of enterprise data - ERP schemas, transactional databases, unstructured documents accumulated over decades.

Here's what nobody tells you in the tutorials:
Chunk size is not a setting. It's an architectural decision.

Too small - you lose context.

Too large - you pollute the retrieval with noise.

The right chunk size depends on your document structure, your query patterns, and your embedding model's context window.

There is no universal default.
Embeddings are only as good as your data quality.

I've seen RAG pipelines hallucinate confidently because the source documents had inconsistent terminology, duplicate entries, and stale records.

Garbage in. Confident garbage out.
Data cleaning is not optional preprocessing.
It's the foundation.

Retrieval is not search.

Most teams treat vector similarity as the complete solution.

It isn't.

Hybrid retrieval - combining dense vector search with BM25 sparse retrieval - consistently outperforms pure semantic search on enterprise data.

Especially for domain-specific terminology, product codes, and structured identifiers.
Reranking changes everything.

Retrieved chunks ranked by embedding similarity alone miss context relevance.

A cross-encoder reranker as a second pass - Cohere Rerank, BGE, FlashRank - dramatically improves answer quality.

This single step is the most underused improvement in production RAG stacks.
The LLM is the last 20% of your pipeline.

I see teams spend 80% of their time prompt engineering the final generation step.

The real leverage is in:
- Document parsing and preprocessing
- Chunking strategy
- Metadata filtering
- Hybrid retrieval
- Reranking
Get those right and almost any capable LLM performs well.

Get those wrong and GPT-4 still gives you garbage.

Context window management is a production concern.

Stuffing every retrieved chunk into the prompt is not a strategy.

Map-reduce summarization, iterative retrieval, and query decomposition become essential at scale.

After 30 years of building enterprise systems I keep seeing the same pattern.
The teams that ship reliable AI aren't the ones with the best models.

They're the ones who treated data architecture with the same rigor they'd apply to any production system.
RAG is not a feature.

It's an engineering discipline.

#RAG #RetrievalAugmentedGeneration #LLM #AIEngineering #VectorDatabase #EnterpriseAI #GenerativeAI #AIArchitecture #NLP #BuildInPublic

[Sridhar Seshan @sri_s3]

@nooriefyi Working on AP Invoice automation

[Sridhar Seshan @sri_s3]

Everyone is building RAG pipelines right now.
Most will fail in production.
Not because RAG is wrong.
Because they skipped the hard parts.
Here's what nobody tells you in the tutorials:

Chunk size is not a setting.
It's an architectural decision.
Too small — you lose context.
Too large — you pollute retrieval with noise.
The right chunk size depends on your document structure, query patterns, and your embedding model's context window.
There is no universal default.

Embeddings are only as good as your data quality.
I've seen RAG pipelines hallucinate confidently because source documents had inconsistent terminology, duplicate entries, and stale records.
Garbage in.
Confident garbage out.
Data cleaning is not optional preprocessing.
It's the foundation.

Retrieval is not search.
Most teams treat vector similarity as the complete solution.
It isn't.
Hybrid retrieval — dense vector search + BM25 sparse retrieval — consistently outperforms pure semantic search on enterprise data.
Especially for domain-specific terminology and structured identifiers.

Reranking changes everything.
Retrieved chunks ranked by embedding similarity alone miss context relevance.
A cross-encoder reranker as a second pass — Cohere Rerank, BGE, FlashRank — dramatically improves answer quality.
Most underused improvement in production RAG stacks.

The LLM is the last 20% of your pipeline.
Most teams spend 80% of their time prompt engineering the final generation step.
Real leverage is in:
— Document parsing
— Chunking strategy
— Metadata filtering
— Hybrid retrieval
— Reranking
Get those right and almost any LLM performs well.

Context window management is a production concern.
Stuffing every retrieved chunk into the prompt is not a strategy.
Map-reduce summarization, iterative retrieval, and query decomposition become essential at scale.

After 30 years building enterprise systems I keep seeing the same pattern.
Teams that ship reliable AI aren't the ones with the best models.
They're the ones who treated data architecture with the same rigor they'd apply to any production system.

RAG is not a feature.
It's an engineering discipline.
Same fundamentals that made ERP implementations succeed or fail — data quality, process mapping, exception handling — apply directly here.
The stack changed.
The discipline didn't.

If you're building RAG on top of enterprise or SMB data and hitting walls in production — happy to compare notes.

What's the hardest part of your pipeline right now? 👇

[Sridhar Seshan @sri_s3]

I'll be honest about something most builders won't say publicly.

Our first version of Ordr AI failed in production.

Not the demo. The demo was smooth.
Production.

A customer called. Said "I want the chicken biryani, extra spicy, no raita, and can you check if the mutton is available today."

The agent froze.

We had built it for clean, predictable inputs.
Real customers don't talk like that.

They pause. They change their mind mid-sentence. They ask questions we never anticipated.

I've seen this exact failure before.

25 years ago. Different context.

A manufacturer went live on a new ERP system.

Worked perfectly in UAT.

Fell apart on day one because real purchase orders looked nothing like the test data.
Same lesson. Different decade.

The gap between "works in testing" and "works in the real world" is where most software projects die.

AI agents are no different.

What saved Ordr AI?

We sat in the restaurant during service.
Watched real orders. Listened to real customers.

Rebuilt the conversation flow around what actually happens — not what we assumed.

The fix wasn't technical.

It was humility.

Go watch your process before you automate it.

Every single time.

#BuildInPublic #VoiceAI #AIAgents #OrdrAI #Lessons #SMB #ArtificialIntelligence

[Sridhar Seshan @sri_s3]

1/ I spent 25 years migrating ERP systems for manufacturing and hotel companies.
Now I build AI voice agents that take restaurant orders.
Here's the uncomfortable truth most AI startups won't tell you: 🧵
2/ Every "AI transformation" I've seen fail in the last year had the same root cause.
Nobody mapped the actual business process first.
They went straight to "let's add AI" — same mistake I watched companies make with ERP rollouts in 2005.
3/ Back then it was: "Let's implement SAP/Sage and figure out our process later."
Now it's: "Let's plug in an AI agent and figure out what it should actually do later."
The tool changed. The mistake didn't.
4/ When I built Ordr AI (a voice ordering agent for a restaurant client), the AI part was maybe 20% of the work.
The other 80%? Understanding how orders actually flow — kitchen timing, menu exceptions, how staff handle edge cases.
5/ Sound familiar? It's the exact same discovery work I used to do before any ERP go-live.
Requirements gathering didn't become obsolete because of AI. It became MORE important.
6/ The difference is speed of iteration.
In my ERP days, a process change meant a dev cycle, testing, a deployment window.
With AI agents, I can adjust the agent's logic same-day based on what we observe.
7/ That's the real unlock for small businesses.
Not "AI replaces your staff."
It's "AI gives small businesses the same process-rigor that used to cost enterprises millions — at a fraction of the cost and time."
8/ The businesses winning with AI right now aren't the ones with the fanciest models.
They're the ones whose builders actually understand operations — inventory, scheduling, customer flow, edge cases.
9/ That's why I think this is actually a great moment for people with deep domain/ops experience to move into AI building.
Your "old" skills — process mapping, systems thinking — are suddenly the differentiator again.
10/ I'm building a few of these (voice ordering, AI receptionists, back-office automation) for SMBs and via Ananta.
If you run a small business and are curious what this could look like for you, happy to chat. 👇

[Sridhar Seshan @sri_s3]

#fable was really great in the results , now it's been suspended or banned for our usage. Is this indirect restriction of power of AI usage. #ai #anthropic #claude

[Sridhar Seshan @sri_s3]

I'm Sridhar Seshan , Founder of Condor Technology Associates

I help entrepreneurs, students & business owners understand technology & innovate smarter.

On this account, I share:
→ Tech trends that matter
→ Innovation strategies that work
→ Real lessons from building a tech company

Follow me if you want to stay ahead of the curve 👉

[Sridhar Seshan @sri_s3]

We started training robots for daily chores

[Vaibhav Sisinty @VaibhavSisinty]

A 25-year-old housewife in Chennai earns ₹250/hour ($3) just by doing her normal housework.

She wears a phone on her head and records herself making coffee, cutting fruit, folding laundry.

These first-person videos get sent to AI companies training humanoid robots to handle real-world tasks. She shoots 90+ clips a day.

Her quote: "Who else will pay you ₹250/hour ($3) an hour just for doing housework?"

She's part of a growing gig economy in India where thousands are doing the same thing, filming everyday life to train the robots of tomorrow.

[Sridhar Seshan @sri_s3]

Stop treating AI like a microwave.

Right now, Google’s AI Studio is helping teams pump out over 1.2 million apps every single week. Billions of tokens are being burned in a massive game of "tokenmaxxing."

By all metrics, we should be living in a golden age of software innovation. So why does the app store feel exactly the same?

Because we fell for the ultimate illusion: Confusing velocity with value.
Everyone is obsessing over the upcoming release of Fable 5 and chasing the shinier, faster, bigger frontier models. But the hard truth is that if you put a microwave on turbo-speed, it still just heats up leftovers.

When you prioritize quantity over quality, AI doesn’t accelerate your team. It just helps them generate technical debt at the speed of light.

The real AI revolution isn't happening in the labs; it's happening in the application.

The true innovators aren't the ones bragging about their token spend. They are the builders, operators, and product minds who realize two things:
- You rarely need a bleeding-edge frontier model to solve a real human problem.
- An AI agent is only as good as the workflow it disrupts.

Amazon’s recent moves have hinted at this exact shift: moving away from the hype of raw compute and toward practical, high-quality execution.

In a world where anyone can generate 10,000 lines of code with a single prompt, the ultimate competitive advantage is taste.

Before you hit deploy on that next AI feature, ask yourself the golden question: "If my name was the only one on this, would I still ship it?"
Let’s stop trying to out-produce the machines and start out-thinking them.

Are we drowning in AI noise, or is this just the messy middle before the real value kicks in? Drop your take below.

#ArtificialIntelligence #TechLeadership #Fable5 #ProductStrategy #SoftwareEngineering #GenerativeAI #Innovation #FutureOfWork

[Sridhar Seshan @sri_s3]

@Berry_cooool Auto

[Jouhatsu | AI Influence Operator @Jouhatsu_ai]

Un ingénieur IA senior chez Microsoft vient de dévoiler comment les équipes de Microsoft créent des agents IA avec Anthropic.

34 minutes de workshop gratuit, directement par l’équipe Microsoft.

Regarde le workshop. Ajoute en signet 🔖

Opus 4.7 + plus de 1 400 outils MCP déjà prêts à l’emploi.

Tu connectes Claude à un agent → tu lui ajoutes des outils → tu déploies en production.

Plus utile que la majorité des formations de vibe-coding vendues 500 $.

[Prajwal Tomar @PrajwalTomar_]

IT'S SO OVER!!! Design agencies are absolutely cooked.

I just connected Codex to Moonchild and generated:

→ Complete design system
→ Every app screen
→ Production-ready code

All in under an HOUR.

Most people are still paying $3k to $10k and waiting WEEKS for Figma handoffs like it's 2023.

This article breaks down the EXACT workflow that's working right now.

[Arish @arishs24]

$4700 to hear your family never made sense to me, we changed that. Pre order now.

[Noisy @noisyb0y1]

Anthropic pays $750,000+ a year for engineers who know how to build LLMs from scratch.

Stanford just released the exact lecture that teaches it - 1 hour 44 minutes, free, straight from CS229.

Bookmark this & give it 2 hours today. It'll teach you more about how ChatGPT & Claude actually work than most people at top AI companies learn in their entire careers.