Search Tech Journey

Find topics, journeys and posts

Long-form self-study

Learning

Every long-form series on dineshblog. Each one is an evergreen path with individual session pages that route out from a hub post. Pick a path. Bookmark the hub. Work through it at your own pace.

80 sessions~160 hrs

Deep Learning & LLMs From Scratch

Zero DL knowledge on day one → your own instruction-tuned 100M-param model, served behind an HTTPS endpoint.

Modules
  • 01Math for DL
  • 02Neural nets in NumPy
  • 03PyTorch fluency
  • 04Regularization + optimization
  • 05Convnets + vision
  • 06RNNs, embeddings
  • 07Transformers from scratch
  • 08Tokenization + scaling laws
  • 09Pretraining your foundation model
  • 10Fine-tuning + alignment (LoRA/QLoRA/DPO)
  • 11Efficient inference (quant / vLLM / spec decoding)
  • 12Multimodal + agents + RAG + capstone
End state
  • Rebuild nanoGPT line by line
  • Pretrain a 100M-param LLM for ~$3
  • SFT + LoRA/QLoRA fine-tune
  • INT4 quantize + serve with vLLM
Audience: You know Python. That's it.
Open the hub post Session index
130 sessions~260 hrs

6-Month Engineering Plan

130 atomic sessions across 12 modules — the whole SW→DE→ML→LLM→MLOps stack in one DAG, spiralling.

Modules
  • 01Setup & Tools
  • 02Python Foundations
  • 03Math Foundations
  • 04DS & Algos
  • 05Databases & SQL
  • 06Data Engineering
  • 07Backend & APIs
  • 08Systems & Infrastructure
  • 09Machine Learning
  • 10LLMs & Modern AI
  • 11MLOps & Production
  • 12System Design + Capstone
End state
  • Full-stack fluency top to bottom
  • Ship 3 capstone projects
  • Read prod code without flinching
  • Interview-ready for L4/L5 IC roles
Audience: Any engineer who wants breadth + depth.
Open the hub post Session index
65 sessions~80 hrs

LeetCode — From Basics to Interview-Ready

65 sessions: never solved a problem before → walking into a FAANG loop with a pattern for everything and a plan for the follow-up.

Modules
  • 01How to read a problem (UMPIRE)
  • 02Complexity + Python fluency
  • 03Arrays, hashing, two pointers
  • 04Sliding window + prefix sums
  • 05Binary search + sorting
  • 06Linked lists, stacks, queues
  • 07Trees + BSTs
  • 08Graphs (BFS/DFS/topo/union-find)
  • 09Heaps, intervals, greedy
  • 10Dynamic programming
  • 11Backtracking + advanced patterns
  • 12Mock loops + company-tagged prep
End state
  • Recognise the pattern before you code
  • Explain complexity out loud, every time
  • Own the follow-up they escalate to
  • Run a timed mock without freezing
Audience: You can write a for-loop. Start here.
Open the hub post Session index
22 sessions~44 hrs

Microsoft Web Stack — Novice to Fluent

22 chapters: .NET runtime → ASP.NET Core → React/TS → Azure → security → testing → pro skills.

Modules
  • 01Developer tooling
  • 02Web fundamentals
  • 03C# language + .NET runtime
  • 04ASP.NET Core
  • 05React + TypeScript
  • 06Frontend build & state
  • 07Auth (Google OAuth PKCE)
  • 08Data + EF Core + Postgres
  • 09Azure App Service + Front Door
  • 10Testing (xunit + Playwright)
  • 11Security + observability
  • 12Shipping workflow
End state
  • Ship a .NET 8 + React app to Azure
  • Google OAuth PKCE, from scratch
  • Postgres migrations that don't lie
  • Front Door + CI/CD wired up
Audience: You know one language. Any language.
Open the hub post Session index
8 sessions~12 hrs

Applied LLMs — Origins to Production

From where LLMs came from, through how attention actually works, to RAG, prompting, audio, medical NLP and shipping on Azure AI Foundry.

Modules
  • 01Where LLMs came from
  • 02How transformers attend
  • 03RAG architecture basics
  • 04Prompting for depth
  • 05Generating audio
  • 06Medical NLP case study
  • 07Stanford CS229 notes
  • 08Azure AI Foundry
End state
  • Explain attention from first principles
  • Design a RAG pipeline
  • Write prompts that hold up
  • Ship an LLM app on Azure
Audience: You've used ChatGPT. Now understand it.
Open the hub post Session index
5 sessions~9 hrs

Data Platform & Orchestration

Kafka, Airflow, Azure Data Explorer and the infrastructure that feeds AI and experimentation at scale.

Modules
  • 01Kafka 101 for ML
  • 02Apache Airflow in depth
  • 03Azure Data Explorer / Kusto
  • 04Data infra for experimentation
  • 05Engineering clarity deep dive
End state
  • Reason about streaming vs batch
  • Orchestrate real pipelines
  • Query at scale with Kusto
  • Design experimentation infra
Audience: You write data pipelines. Level up.
Open the hub post Session index
3 sessions~6 hrs

Recommendation Systems From Scratch

Design a recommender from first signals, build a worked example, then dissect a production engine.

Modules
  • 01Design from first signals
  • 02A worked build
  • 03Production engine teardown
End state
  • Model user/item signals
  • Build candidate + ranking stages
  • Read a real recsys architecture
Audience: You know some ML. Build the classic system.
Open the hub post Session index
4 sessions~6 hrs

Ship It — Deploy, Package, Sustain

Scripts to APIs, Docker on Azure, Windows apps to the store — plus the learning system that keeps it all sustainable.

Modules
  • 01Scripts → API on Azure
  • 02Docker deployment on Azure
  • 03Windows app → store
  • 04A learning system that sticks
End state
  • Deploy a script as an API
  • Containerise + ship on Azure
  • Publish a desktop app
  • Build a durable study loop
Audience: You can build. Learn to ship + sustain.
Open the hub post Session index
20 sessions~40 hrs

Designing for Scale — Requirements to Review

Constraints and estimation first, then eight classic designs, then the data and ML systems underneath — closing with two end-to-end design reviews.

Modules
  • 01Requirements → architecture
  • 02Estimation that constrains
  • 03Storage decision tree
  • 04CAP and consistency
  • 05Six classic designs
  • 06Real-time analytics + CDC
  • 07Lakehouse and metering
  • 08Feature store + model serving
  • 09RAG and eval gates
  • 10Two full design reviews
End state
  • Derive an architecture from constraints
  • Estimate well enough to rule options out
  • Defend a storage choice under questioning
  • Run an end-to-end design review
Audience: You can build a service. Now defend it at scale.
Open the hub post Session index

How to use this page

  • · Pick one series. Don't series-hop — depth beats breadth here.
  • · Open the hub post. It has the full session list with click-through.
  • · Read one session end to end (~2 hrs) — including the recall questions at the bottom.
  • · Come back the next day. Every session ends with a “bring back tomorrow” block for a reason.