Harsh Arora

Building with confidence, growing through challenges.
Gen Ai • NEXT• MERN • DevOps • React Native

[Harsh Arora @AroraHarsh17]

@AramideOyekunle Interested

[Harsh Arora @AroraHarsh17]

@Lawalsegun_ Hi

[Harsh Arora @AroraHarsh17]

@HRRonald001 Full stack dev

[Harsh Arora @AroraHarsh17]

@MrCharlesky Interested

[Harsh Arora @AroraHarsh17]

@JobFound5 Interested

[Harsh Arora @AroraHarsh17]

@JobFound5 Interested

[Harsh Arora @AroraHarsh17]

@irnofic Interested

[Harsh Arora @AroraHarsh17]

@connoratlunon https://t.co/JxERA3YNVB

[Harsh Arora @AroraHarsh17]

@nitesh_singh5 Interested

[Harsh Arora @AroraHarsh17]

@nitesh_singh5 Interested

[Harsh Arora @AroraHarsh17]

🚀 Built an Agentic AI WhatsApp bot.
User message → classify → router → specialized agents (Code, Content, Math) → review → reply.
Each agent uses different LLM configs & temperature tuning for task-specific responses.
Built using FastAPI + LangChain.

[Harsh Arora @AroraHarsh17]

Built a fully async RAG system 🚀

⚡ FastAPI
🔁 Redis Queues + Workers
🧠 Qdrant
🤖 OpenAI

Flow:
User → /chat → Redis Queue → Worker (embed → retrieve → generate) → Store → /result/{job_id}

No blocking. Scalable. Production-ready.

GitHub: https://t.co/p7xQNo81a8

[Harsh Arora @AroraHarsh17]

Why It Matters?

Without RAG → LLM guesses
With RAG → LLM answers using context.

[Harsh Arora @AroraHarsh17]

RAG is how you turn LLMs from guessing machines into knowledge-powered systems.

Here’s how RAG actually works (in simple terms) 🧵👇

[Harsh Arora @AroraHarsh17]

Phase 2: Retrieval
• User sends query
• Convert query into embedding
• Similarity search
• Send relevant chunks + query to LLM

[Harsh Arora @AroraHarsh17]

Built my first CLI Weather Agent 🌦️
Used tool calling to fetch real-time weather data.
Biggest learnings:
• Prompt quality matters more than I thought
• Explored prompting styles: Direct, One-shot, Few-shot & Chain of Thought
• Better prompts = smarter agents

[Harsh Arora @AroraHarsh17]

One step toward Gen AI 🚀

asyncio in python feels a lot like async/await in JS called corutine in python same event-loop magic.
Big takeaway: thread pools for I/O-bound work, process pools for CPU-heavy tasks, and daemon threads for background jobs that shouldn’t block shutdown.

[Harsh Arora @AroraHarsh17]

Goals for this year :-

1. AWS Cloud Practitioner Certification
2. Learn remaining devops conceps like Kubernetes, Grafana etc
3. Learn Gen Ai concepts
4. Build more while learning
5. Grab better opportunities