CEO @ HiCounselor | 2X O-1 Visa Holder | Ex-Deloitte, PwC | Building a Community to Help Job Seekers with Referrals

I came to the US in 2014 for a master's degree at Duke University.

Post-graduation, I was fortunate to start my career as a Tech Consultant at PwC and then with Deloitte in San Francisco. 

However, I always wanted to utilize my tech skills to start my own business. 

In 2020, I decided to leave my job and launched HiCounselor in San Francisco.

The decision wasn't easy as I had many open questions, concerns about immigration, and fear of failure. 

Fortunately, the decision turned out to be the right one. 
✅ Generated millions in revenue
✅ Raised VC funds from top investors in Silicon Valley
✅ Employ more than 40 people in the US, Canada, and India.

👉 These four years have taught me a lot about starting a company as an immigrant in the US, and I want to share this journey with every technical professional who wants to make side income or become a full-time entrepreneur.

✉️ Join my newsletter (https://lnkd.in/gPBziwZ3), where I share weekly actionable tips. 


👉  If you are not ready to leave your full-time job at this time, that's okay.
You can join hicounselor.com/mentor and make a side income by sharing your technical expertise. 

If you need any help, feel free to message me or comment on my posts. I will respond as soon as possible.

Aditya Sharma

𝗪𝗵𝘆 𝗱𝗼 𝗺𝗼𝗱𝗲𝗹𝘀 𝘀𝘁𝗶𝗹𝗹 𝗵𝗮𝗹𝗹𝘂𝗰𝗶𝗻𝗮𝘁𝗲? 🤔
Most hallucinations are actually RAG stack failures caused by broken context.

When retrieval is weak, the model does exactly what it was trained to do. It completes the pattern using the wrong information.

This is why understanding the full RAG stack matters.

𝗛𝗲𝗿𝗲’𝘀 𝘄𝗵𝗮𝘁 𝗮 𝗰𝗼𝗺𝗽𝗹𝗲𝘁𝗲 𝗢𝗽𝗲𝗻-𝗦𝗼𝘂𝗿𝗰𝗲 𝗥𝗔𝗚 𝗦𝘁𝗮𝗰𝗸 𝗹𝗼𝗼𝗸𝘀 𝗹𝗶𝗸𝗲 👇

1. Ingest & Data Processing
↳ Collect, clean, and structure raw data
↳ Poor ingestion leads to poor retrieval
Tools: Kubeflow, Apache Airflow, Apache NiFi, LangChain Loaders, Haystack Pipelines, OpenSearch

2. Embedding Models
↳ Convert text into semantic meaning
↳ Determines how well the system understands similarity
Tools: HuggingFace Transformers, Sentence Transformers, JinaAI, Cognita, Nomic, LLMWare

3. Vector Databases
↳ Store embeddings efficiently for fast lookup
↳ Enables scalable semantic search
Tools: Milvus, Weaviate, PgVector, Chroma, Qdrant

4. Retrieval & Ranking
↳ Finds the most relevant context for each query
↳ Weak retrieval = confident hallucinations
Tools: FAISS, Weaviate, JinaAI Rerankers, Haystack Retrievers

5. LLM Frameworks & Orchestration
↳ Connect pipelines, tools, and workflows
↳ Ensure reliability end to end
Tools: LangChain, LlamaIndex, Haystack, CrewAI, HuggingFace

6. LLMs (The Responder)
↳ Generate answers grounded in retrieved context
↳ Model size matters less than context quality
Models: LLaMA, Mistral, Phi-2, DeepSeek, Qwen, Gemma

7. Frontend Layer
↳ Where users interact with intelligence
↳ Trust is built at the interface
Tools: NextJS, Streamlit, VueJS, SvelteKit

🔥 𝗞𝗲𝘆 𝘁𝗮𝗸𝗲𝗮𝘄𝗮𝘆:
Most hallucinations are not caused by bad models.
They’re caused by bad context.

📌 𝗚𝗲𝘁 𝗔𝗜 𝗔𝗴𝗲𝗻𝘁 𝗖𝗼𝘂𝗿𝘀𝗲𝘀 𝗟𝗶𝘀𝘁 (𝗳𝗿𝗲𝗲): https://lnkd.in/gHRjrXQG

👉 Follow me Aditya Sharma for more and 🔄 Repost this to help others use AI

#AI #RAG #OpenSource

Helping Top 1% AI Talent Land Role at Elite AI Companies | Turning Ideas into Real Businesses Using AI | 160k+ Followers | Ex-Deloitte & PwC

Everyone is talking about Agentic AI…
but almost no one tells you when to use which tool.

So I made this simple breakdown 👇
Top 10 Agentic AI tools and where they actually shine.

Most people jump straight into building agents
→ then struggle with chaos, loops, or over-engineering.


𝗥𝗲𝗮𝗹𝗶𝘁𝘆 𝗶𝘀 𝘀𝗶𝗺𝗽𝗹𝗲:
↳ Some tools are great for automation
↳ Some for multi-agent collaboration
↳ Some for RAG + memory
↳ Some for enterprise workflows
↳ Some just for quick demos


𝗛𝗲𝗿𝗲’𝘀 𝗵𝗼𝘄 𝘁𝗼 𝗰𝗵𝗼𝗼𝘀𝗲 👇

> 𝗻𝟴𝗻 → Use this when you want fast workflow automation, lots of app integrations, or self-hosted control with minimal code.

> 𝗔𝘂𝘁𝗼𝗚𝗲𝗻 → Use this when multiple AI agents need to talk, debate, and collaborate on complex problems.

> 𝗟𝗮𝗻𝗴𝗖𝗵𝗮𝗶𝗻 → Use this when you’re building real AI products with LLMs, tools, memory, or advanced RAG pipelines.

> 𝗠𝗮𝗸𝗲.𝗰𝗼𝗺 → Use this when non-technical teams need visual, no-code automation across SaaS tools.

> 𝗖𝗿𝗲𝘄𝗔𝗜 → Use this when agents need clear roles (researcher, planner, executor) and structured teamwork.

> 𝗙𝗹𝗼𝘄𝗶𝘀𝗲 → Use this when you want drag-and-drop AI agents or quick RAG prototypes.

> 𝗟𝗮𝗻𝗴𝗚𝗿𝗮𝗽𝗵 → Use this when agents must plan, remember, loop, and handle complex state-based logic.

> 𝗢𝗽𝗲𝗻𝗔𝗜 𝗔𝗴𝗲𝗻𝘁𝗶𝗰 𝗦𝘁𝗮𝗰𝗸 → Use this when you need quick demos with built-in tools, memory, and function calling.

> 𝗟𝗹𝗮𝗺𝗮𝗜𝗻𝗱𝗲𝘅 → Use this when your AI must deeply understand and reason over private or structured data.

> 𝗦𝗲𝗺𝗮𝗻𝘁𝗶𝗰 𝗞𝗲𝗿𝗻𝗲𝗹 → Use this when you’re building enterprise-grade, scalable, memory-aware AI systems.

The biggest mistake people make?
Using enterprise tools for prototypes
or prototype tools for production

📌 𝗚𝗲𝘁 𝗔𝗜 𝗔𝗴𝗲𝗻𝘁 𝗖𝗼𝘂𝗿𝘀𝗲𝘀 𝗟𝗶𝘀𝘁 (𝗳𝗿𝗲𝗲): https://lnkd.in/gHRjrXQG

👉 Follow me Aditya Sharma for more and 🔄 Repost this to help others.

#AgenticAI #AIAgents #AIEngineering

AI Agents are quickly becoming the foundation of modern automation, products, and AI-powered businesses.

In this session of the AI for Everyone series, this beginner-friendly live workshop will help you understand how AI agents actually work, how they learn, and how you can start using them confidently even if you are new to AI.

This is a conceptual + practical fundamentals session, designed to remove confusion and build strong foundations.

What you’ll learn in this 30-minutes live session:

⚡ AI Agent Fundamentals
Understand what AI agents are, how they differ from AI tools, and where they are used.

🤖 How AI Agents Learn and Make Decisions
Learn the core concepts behind memory, tools, reasoning, and automation.

🛠 Real Examples and Workflows
See how AI agents operate in real scenarios across business and personal workflows.

🌟 Getting Started the Right Way
Learn how beginners, professionals, and founders can start building and using AI agents correctly.

Whether you are a student, beginner, recruiter, founder, tech professional, or simply curious about AI, this session will give you clarity and confidence to move forward.

✨ Free to attend. Register now to save your spot.

Session 16: Fundamentals of Learning AI Agents

Healthcare-focused Data Analyst with 2+ years of experience using SQL, Python, Excel, and ETL (Alteryx, KNIME) to analyze clinical and operational data. Skilled in data cleaning, ETL, and reporting to support patient care, triage accuracy, and quality metrics. Proven success building dashboards and driving data-driven decisions in healthcare.

Career Highlights:
- Cut patient care response time by 25%, reducing intervention delays from 8 to 6 hours a role-specific decision-support dashboard based on lab response workflows.
- Identified high-risk communities with 95% accuracy across 330M+ residents by surfacing health access disparities and visualizing results in Power BI.
- Increased student retention by 25% across 800+ undergraduates by developing early alert models and academic risk dashboards in Tableau.

I get excited about opportunities where I can bridge the gap between data and human outcomes, especially roles that combine healthcare impact, technical depth, and close collaboration with care teams. I thrive in environments where I can ask the right questions, build scalable tools, and make data intuitive for those on the frontlines of care.

Outside of work, I love exploring new places and experiencing how different communities live, eat, and connect. Travel keeps me curious and open-minded. I’m also passionate about animals and regularly volunteer with local shelters, it’s one of the most grounding and joyful parts of my life. Whether I’m planning my next trip or spending time with dogs in need of a home, I’m happiest when I’m connecting with the world in a meaningful way.

Sriram Srinivasan

📦 𝗣𝗿𝗼𝗷𝗲𝗰𝘁 𝟰 (𝗣𝗮𝗿𝘁 𝟯): 𝗧𝗲𝗮𝗰𝗵𝗶𝗻𝗴 𝗠𝘆 𝗔𝗴𝗲𝗻𝘁 𝘁𝗼 𝗥𝗲𝗮𝗱 𝗔𝗺𝗮𝘇𝗼𝗻 𝗮𝘁 𝗦𝗰𝗮𝗹𝗲

As my AI agent from Wayfair's Externship evolves into a marketing intelligence system, one source remains unavoidable: Amazon.

Not because it’s clean or simple. Because it’s where real consumer demand leaves fingerprints.

This part of Project 4 focuses on how my agent ingests Amazon data in a way that’s robust and signal-driven, even when the input varies.

🧠 𝗧𝗵𝗲 𝗕𝗶𝗴 𝗜𝗱𝗲𝗮
Amazon doesn’t come in one neat format.

Sometimes you have direct product links Other times, collection pages or search results.

So instead of assuming how Amazon data will arrive, I designed the workflow to adapt.

🧩 𝗧𝘄𝗼 𝗣𝗮𝘁𝗵𝘀, 𝗢𝗻𝗲 𝗚𝗼𝗮𝗹
At a high level, the workflow does two things in parallel:

𝟭. 𝗣𝗿𝗼𝗱𝘂𝗰𝘁-𝗟𝗲𝘃𝗲𝗹 𝗦𝗰𝗿𝗮𝗽𝗶𝗻𝗴
When a direct Amazon product URL is available, the agent:
 • Fetches the product page
 • Extracts core signals like price, materials, patterns, and feature highlights

This gives precise, SKU-level insight into what’s selling and how it’s positioned.

𝟮. 𝗖𝗼𝗹𝗹𝗲𝗰𝘁𝗶𝗼𝗻-𝗟𝗲𝘃𝗲𝗹 𝗘𝘅𝗽𝗮𝗻𝘀𝗶𝗼𝗻
When the input is broader, the agent:
 • Pulls the full HTML
 • Extracts all product URLs
 • Cleans and standardizes them
 • Splits them into individual items
 • Scrapes each product one by one

This is how the agent moves from a few examples to market-level visibility.

🔁 𝗪𝗵𝘆 𝗜𝘁 𝗟𝗼𝗼𝗸𝘀 “𝗕𝘂𝘀𝘆” (𝗢𝗻 𝗣𝘂𝗿𝗽𝗼𝘀𝗲)
The multiple branches, waits, and splits you see in the screenshot aren’t accidental.

They exist to handle:
• Rate limits
• Partial URLs
• Duplicate products
• Inconsistent page structures

In other words, this isn’t a scraper that works when conditions are perfect. It’s a system designed to keep working when they’re not!

💡 𝗪𝗵𝘆 𝗧𝗵𝗶𝘀 𝗠𝗮𝘁𝘁𝗲𝗿𝘀 𝗳𝗼𝗿 𝗠𝗮𝗿𝗸𝗲𝘁𝗶𝗻𝗴 𝗜𝗻𝘁𝗲𝗹𝗹𝗶𝗴𝗲𝗻𝗰𝗲
Amazon isn’t just a retailer. It’s a live demand signal.

By structuring Amazon data this way, the agent can later answer questions like:
 • Which styles are gaining traction right now?
 • What price ranges are getting the most engagement?
 • Which materials and features are repeatedly emphasized?
 • How competitor messaging shifts across price tiers?

This turns Amazon from a chaotic marketplace into a signal engine.

🧠 𝗧𝗵𝗲 𝗕𝗶𝗴𝗴𝗲𝗿 𝗣𝗶𝗰𝘁𝘂𝗿𝗲
This workflow isn’t about scraping more data.

It’s about giving the agent situational awareness, so when it generates insights, ideas, or recommendations, they’re grounded in what consumers are actually responding to.

And that’s what Project 4 is really about: turning raw market noise into usable marketing intelligence.

📸 𝗔𝘁𝘁𝗮𝗰𝗵𝗲𝗱: Amazon scraping workflow showing both product-level and collection-level paths.

Follow Sriram for bite-sized GenAI, automation, and data science breakdowns.

#AIAgents #Amazon #Wayfair #MarketingIntelligence #n8n #GenerativeAI #Automation #BuildingInPublic

Awarded Top Performer | Healthcare Data Analyst | Python, SQL, Power BI, Cloud | Building dashboards that reduce delays by 25% & support equitable healthcare decisions

taplio

The Battle

Aditya Sharma

Sriram Srinivasan

Followers

Posts

Engagements

Likes

Performance ScoreAditya Sharma's LinkedIn Performance

Performance ScoreSriram Srinivasan's LinkedIn Performance

Ready to grow your LinkedIn presence?

taplio

The Battle

Aditya Sharma

Sriram Srinivasan

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);}Aditya Sharma's LinkedIn Performance

Performance ScoreSriram Srinivasan's LinkedIn Performance

Ready to grow your LinkedIn presence?

Performance ScoreAditya Sharma's LinkedIn Performance