Shubham Sarpate

Full Stack Developer| Learning Generative AI 🚀| Curious tech enthusiast| IIITR'25

[Shubham Sarpate @ShubhamSr06]

Bura mera waqt ho sakta hai, bura main insaan nahin.
Meri ghaltiyan maaf karo kyunki main Bhagwan nahin.
Mujhe itna beqadar bhi kyun karte ho,
Main to phir bhi insaan hoon, ghar mein pada koi samaan nahin.

[Shubham Sarpate @ShubhamSr06]

Big takeaway:

Embedding models create vector representations of data, while vector databases store and retrieve those vectors efficiently.

Both are essential for building fast and accurate RAG applications.

Day 15/30 ✅

#GenerativeAI #RAG #LangChain #VectorDB #BuildInPublic

[Shubham Sarpate @ShubhamSr06]

Day 15 of my Generative AI journey 🚀

Today I moved from understanding the theory behind RAG to learning how it's implemented in practice.

Here’s what I learned 👇

[Shubham Sarpate @ShubhamSr06]

• The difference between embedding models and vector databases
• Popular embedding models like OpenAI, Gemini, BERT, and all-MiniLM-L6-v2
• Popular vector databases like Chroma, FAISS, Qdrant, Pinecone, and Weaviate

[Shubham Sarpate @ShubhamSr06]

अपने आप के भी पीछे खड़ा हूँ मैं,
ज़िंदगी, कितने धीरे चला हूँ मैं।

[Shubham Sarpate @ShubhamSr06]

Big takeaway:

Modern AI systems don't rely on a single model.

RAG combines embedding models for retrieval with generative models for answering questions—each model does what it's best at.

Day 14/30 ✅

#GenerativeAI #LLM #Embeddings #RAG #BuildInPublic

[Shubham Sarpate @ShubhamSr06]

Day 14 of my Generative AI journey 🚀

Today I focused on understanding how LLMs represent language internally and why different models have different roles in AI systems.

Here’s what I learned 👇

[Shubham Sarpate @ShubhamSr06]

• Auto-Encoding models like BERT and how they learn to understand text
• Auto-Regressive models and why they excel at text generation
• Understanding vs Generation: different training objectives, different use cases

[Shubham Sarpate @ShubhamSr06]

Big takeaway:

RAG isn't about making an LLM smarter—it's about giving it the right information at the right time.

Day 13/30 ✅

#GenerativeAI #RAG #LLM #AIEngineering #BuildInPublic

[Shubham Sarpate @ShubhamSr06]

Day 13 of my Generative AI journey 🚀

Today I explored one of the most important concepts in modern AI applications: Retrieval-Augmented Generation (RAG).

Here’s what I learned 👇

[Shubham Sarpate @ShubhamSr06]

• Why RAG doesn't train the LLM
• RAG vs Fine-tuning
• Advantages, limitations, and common real-world use cases
• Why RAG has become a standard approach for building AI applications

[Shubham Sarpate @ShubhamSr06]

Big takeaway:

Evaluating AI models isn't just about technical metrics.

The best model is the one that delivers the best outcome for your specific use case, balancing quality, speed, cost, and user experience.

Day 12/30 ✅

#GenerativeAI #LLM #AIEngineering #BuildInPublic

[Shubham Sarpate @ShubhamSr06]

Day 12 of my Generative AI journey 🚀

Today was more hands-on as I explored how to evaluate and compare frontier LLMs for real-world development tasks.

Here’s what I learned and built 👇

[Shubham Sarpate @ShubhamSr06]

• Built a Gradio UI to compare Python-to-C++ code conversion across multiple models
• Explored how a simple interface can make model evaluation much easier

[Shubham Sarpate @ShubhamSr06]

Big takeaway:

Choosing an LLM isn't just about picking the model with the highest benchmark score.

The best model depends on the task, cost, latency, and the problem you're trying to solve.

Day 11/30 ✅

#GenerativeAI #LLM #AIEngineering #BuildInPublic

[Shubham Sarpate @ShubhamSr06]

Day 11 of my Generative AI journey 🚀

Today I shifted my focus from understanding how LLMs work to understanding how to evaluate and choose the right model for real-world applications.

Here’s what I learned 👇

[Shubham Sarpate @ShubhamSr06]

• Why benchmarks aren't perfect—data contamination and overfitting can affect results
• How to compare models using leaderboards like Artificial Analysis and Hugging Face
• Commercial AI use cases: automation, augmentation, and Agentic AI