Petroleum From Scratch

📍Regression Math to LLM Math - Connecting the dots. Every model at the core of it is just y = f(X) and the f keeps getting prettier. 1. f being a linear vector of weights and X being row wise inputs, you’re in the Linear Regression regime. 2. f being a recursive matrix multiplication of weight matrices and sandwiched by non linear activation functions, X being structured matrix inputs, you’re in the deep neural network regime. 3. f being a transformer stack with key query and value interaction formulations in self and cross attention modes with other ingredients like position encoding and normalizations followed by a softmax, you’re in the GenAI LLM modelling regime with a multi class classification way of things i.e. the model classifying the output class from the vocab space. Pardon me for the oversimplification. But I hope it makes sense for some. Happy Learning! No hash-tags.

Wrote an End-To-End #RAG Pipeline to build a custom #ReservoirGPT RAG: Retrieval-Augmented Generation - A technique that can create an AI-Enabled Tool based on your custom documents that can act as your custom GPT, answering queries using the combined powers of your document database and Universal Language Models. - For this (Open Source) Project, I have used important Reservoir Engineering documents (from the web) and fine-tuned a RAG pipeline on them, so as to leverage a GPT Tool that can act as an AI based Reservoir Assistant for me. - This can further extend to #Wells, #Drilling etc. Particulars of this Pipeline. 1. PDF to Text Extraction: Marker, PyMuPDF etc. 2. Max Context Length: 400-450 Tokens 3. Embedding Framework: mx-bai-large-v1 ( 🤗 ) 4. Semantic Search: Explored Cosine similarity and Normalized Dot-Products along with Torch tools to sort max relevance. 5. LLM Used : Google/Gemma-2B-it (Instruction Tuned). Additional Features: - Wrote functions to wrap the text with adequate cushioning to make prompts consistently put. - Added a bit of prompt engineering layer by formatting the user defined prompt with an Example-Answer kind of formatting before the prompt is fed to the LLM. Limitations: 1. No Finetuning has been done to the LLM Yet. 2. Vector Database/Vector Indexing/Vector Search yet to be implemented as the small-ish nature of this project did not demand it. ------------------------------------ Results Demo: User Query: "What are the sedimentary rocks, and what is their porosity?" ReservoirSME.ai (RAG Pipeline) Answer: **Sedimentary rocks** are formed through the accumulation and compaction of sediments over time. These sediments can be deposited in various ways, including: * **River deposition:** Sediment is transported by water from land to a body of water, such as a river or ocean. * **Aeolian deposition:** Wind can transport and deposit sediment over long distances. * **Volcanic ash deposition:** Ash from volcanic eruptions can be deposited as fine-grained sediment. * **Marine deposition:** Sediments can be deposited directly from the sea, such as through the accumulation of fine-grained particles. **Porosity** is a measure of how much space is occupied by pores or air within a rock. The porosity of sedimentary rocks can vary widely, depending on the type of rock, the environment in which it was formed, and the presence of fossils. **Factors affecting porosity:** * **Grain size:** The porosity of sedimentary rocks is typically higher for rocks with a higher proportion of smaller particles, such as sand and silt. * **Interconnectivity:** The porosity of sedimentary rocks is also influenced by the degree of connectivity between pores. -------------------------- Much better than CTRL + F for sure! Best Part?: Ran it all on my Local GPU. Worst Part?: Ran it all on my Local GPU. #OpenAi #LargeLanguageModels #GenerativeAI #GenAI #NVIDIA #GPU #Python

📍If I had to write a #Transformer down as a formula/expression for my old-school heart, here’s how I would do it. #generativeai #llms #GPT

Aramco Digital is on the hunt for a visionary Data Center Engineering Director to spearhead the design and deployment of its groundbreaking next-generation AI data centers. This high-impact role will be at the forefront of revolutionizing the company’s digital and AI infrastructure, shaping the future of cutting-edge technologies that will power Aramco's bold digital transformation. The ideal candidate will be a trailblazer in data center innovation, with deep expertise in cutting-edge cooling technologies, ensuring energy efficiency and sustainability. With a proven track record of building world-class, scalable, and ultra-efficient AI systems, this is a unique opportunity to lead transformative projects that will redefine the future of energy and technology at one of the world’s most influential companies.

📚 From Zero to Non-Zero in #Python programming - a course that I taught at IIT-B. If you've never typed a line of code but you want to, you see the value, you want to make an impact and you want to get on speed real quick- This course might be a good quick catalyst for you! If you do learn a thing or two, do drop a message to me in the chat, and I'll be happy to connect and learn from you as well! Happy Learning!

📍7000.00000 Subscribers. A good number, a good achievement, this. Proud. Happy. Motivated to share a lot more. Petroleum From Scratch team thanks all of you. Happy Learning!

“The world needs CCS to be successful—more than 100 times more successful than it is now, to be exact.” The question is: Can carbon storage succeed without petroleum engineering? Carbon capture and storage (CCS) is top of mind for many, but so far, the progress of advancement is modest at best. Why? Effective CCS requires deep knowledge of the subsurface, which is notoriously complex. Whether you choose to repurpose a deep saline aquifer or depleted oil and gas reservoir, here are the biggest questions you need to answer before moving forward with CCS: 🔵 Capacity: How much CO₂ can be stored underground? For saline aquifers, storage is a function of pore space, salinity, mineralogy, and geomechanics. For depleted reservoirs, storage capacity is more straightforward. 🔵 Injectivity: How fast can CO₂ be injected safely? To maximize injectivity while ensuring safety, factors like reservoir pressure and geomechanics need special consideration. Injection tests provide invaluable understanding. 🔵 Containment: How can we ensure CO₂ is stored safely and permanently with no leakage? Due to long production periods, the well integrity and conditions of depleted reservoirs must be assessed, while caprock integrity and potential CO₂ leak paths need consideration in saline aquifers. We still have quite a journey ahead of us before reaching the full potential of CCS. But one thing we know for sure is that the path forward starts with asking the right questions. Full article from Kamal Jansen our CCS Technical Director, now available: https://shorturl.at/HJOT6