siddddhesh's profile picture. ๐Ÿง‘โ€๐Ÿ’ปAI Engineer @IBM | โœ๏ธ Tech Writer | Passionate about teaching & learning | Exploring AI, productivity, and writing ๐ŸŒฑ one insight at a time.

Sid ๐Ÿš€

@siddddhesh

๐Ÿง‘โ€๐Ÿ’ปAI Engineer @IBM | โœ๏ธ Tech Writer | Passionate about teaching & learning | Exploring AI, productivity, and writing ๐ŸŒฑ one insight at a time.

Agentic AI isnโ€™t sci-fi, itโ€™s here. Think: - Auto-ticket resolution - Self-healing infrastructure - Autonomous data enrichment The leap? Moving from reactive to proactive AI systems.


Some of my best AI projects werenโ€™t just cod, they were collaboration Pairing with design, ops, and domain experts turned โ€˜good modelsโ€™ into impactful products


Not all charts belong in Matplotlib. ๐Ÿ“Š Plotly โ†’ interactive dashboards ๐Ÿ“Š Seaborn โ†’ quick EDA visuals ๐Ÿ“Š Altair โ†’ declarative charting Good visuals = better decisions


Slow Python code? Profile before you optimize import cProfile cProfile .run('my_function()') Guessing the bottleneck is slower than finding it


Data Scientists should learn systems design. Not to replace engineers, but to build AI that survives prod.


Writing tech blogs = explaining to your past self. - Keep intros short - Show working code early - Wrap with โ€˜next stepsโ€™ Your reader came for solutions, not suspense.


If your experiment isn't reproducible, it didn't happen. ๐Ÿ’ก Use fixed random seeds ๐Ÿ’ก Version control datasets ๐Ÿ’ก Log env + library versions ML experiments = science, not art.


Feature engineering > fancy models. 90% of ML lift comes from: - Scaling numeric vars correctly - Encoding categories wisely - Creating domain-specific ratios The best Kaggle winners? Feature wizards. ๐Ÿช„


Sid ๐Ÿš€ ๋‹˜์ด ์žฌ๊ฒŒ์‹œํ•จ

Follow these to upgrade your timeline: Markets: @QuasarMarkets Finance: @qmbigbeat Investing: @_parrotfinance PE: @Nick4Pillars Coaching: @WealthAthlete Attorney: @john_jacob Business: @theonenicka Mindset: @MattHunterCoach Leadership: @leadwithchad


LLMs are expensive Not just in $$: - Latency (affects UX) - Carbon footprint - Prompt engineering overhead Smaller + smarter models might beat โ€˜biggerโ€™ in real-world apps.


Letโ€™s do a Data Science challenge: Given a dataset with 1M rows + 50 cols, find the 3 fastest ways to calculate median of col X in Python. Post your code โฌ‡๏ธ


Data Scientists, your go-to notebook tool? ๐Ÿ“Š 1๏ธโƒฃ JupyterLab 2๏ธโƒฃ VSCode + Jupyter Ext 3๏ธโƒฃ Deepnote 4๏ธโƒฃ Other (comment below!)


Writing docs for developers? Think API-first: - Start with usage examples - Show common pitfalls - Keep sentences under 20 words Docs are a dev tool, not a novel


Sid ๐Ÿš€ ๋‹˜์ด ์žฌ๊ฒŒ์‹œํ•จ

Kolhapur has triumphed Pune in terms of Ganapati idols


Reading Research papers โ‰  skimming formulas Breakdown method: 1๏ธโƒฃ Skim intro & abstract 2๏ธโƒฃ Identify problem & baseline 3๏ธโƒฃ Understand method section 4๏ธโƒฃ Jump to results & limitations Turn 15 pages โ†’ 15 mins


Sid ๐Ÿš€ ๋‹˜์ด ์žฌ๊ฒŒ์‹œํ•จ

Top AI Papers of The Week (August 4-10): - CoAct-1 - ReaGAN - Agentic Web - Seed Diffusion - Efficient Agents - A Taxonomy of Hallucinations - Unified Retrieval Agent for AI Search Read on for more:


LLMs are moving from 'general chatbots' โ†’ domain-specialized copilots. Why? โ€ข Cost optimization โ€ข Accuracy in niche vocab โ€ข Easier eval metrics Enterprise AI in 2025 = smaller, smarter, task-focused models.


If you're not reading regularly, how can you expect to write well? You don't have to constantly be *writing* to improve. Reading is another form of practice; sort of like reviewing game tapes. You learn from studying others' approaches.


programming rule: to learn fast - fail fast.


You are not an imposter Writing makes you a writer


Loading...

Something went wrong.


Something went wrong.