Aman Kumar Srivastava

Building Agentic AI Systems for Business Automation | Data Engineer

[Aman Kumar Srivastava @botz_coder]

@visonmilan let's link up 🔥

[Aman Kumar Srivastava @botz_coder]

33 followers. Not a lot, but every single one matters to me. Building in public, aiming for 100 next 🙏

If you're building, learning, coding, or growing in tech

Say hi & let's connect 👋

[Aman Kumar Srivastava @botz_coder]

@atulmishra1996 Sup bro, let's build together 🔥

[Aman Kumar Srivastava @botz_coder]

@windowmemory Following back! Let's grow together 🙌

[Aman Kumar Srivastava @botz_coder]

@MarkoCirix Let's connect

[Aman Kumar Srivastava @botz_coder]

@clawdreyai Done bro

[Aman Kumar Srivastava @botz_coder]

@Hey_Sakshii Done

[Aman Kumar Srivastava @botz_coder]

@mathiasvrb Thanks bro

[Aman Kumar Srivastava @botz_coder]

@IndieDevFire Done

[Aman Kumar Srivastava @botz_coder]

For people who want to understand each question, here is the one-liner for them

Tokenization + hallucination fixes: Tokenization = breaking text into small pieces (words/sub-words) so the model can read it. Hallucination fixes = giving the model real documents to read from (RAG), keeping answers less "creative" (low temperature), and double-checking facts after generation.
LLM usage in apps (chatbots): You take the raw LLM and add memory (so it remembers the chat) and instructions (system prompt) so it behaves like a proper chatbot instead of just answering once.
LangChain (chains vs graphs): Chains = do step 1, then step 2, then step 3, in a fixed order. Graphs = the flow can branch, loop back, or skip steps based on what happens — more flexible, like a flowchart.
Prompt engineering: Carefully writing your instructions/examples so the AI gives you the answer you actually want.
Transformers + embeddings: Transformers are the AI architecture that looks at all words in a sentence together and figures out which words matter to which. Embeddings are just numbers (vectors) that represent the meaning of a word/sentence so similar things end up with similar numbers.
Temp vs top-k vs top-p: All three control how "random" or "safe" the AI's word choices are. Temperature = overall randomness dial (low = safe, high = random). Top-k = only pick from the top K most likely next words. Top-p = only pick from words that together make up P% probability (adapts automatically).
Chunking strategies: Before feeding documents to the AI, you cut them into smaller pieces so it's easier to search and fits in the AI's memory limit.
Activation functions (transformers): A math function inside the model that decides how much a neuron "fires." Transformers mostly use GELU (or similar) to add non-linearity so the model can learn complex patterns.
Gradient boosting: A method where you build many small decision trees one after another, each one fixing the mistakes of the last one, to get a really accurate final prediction.
MCP/tools integration: A standard way for an AI model to call outside tools (like search, calculator, database) so it's not limited to just its own knowledge.
DL vs GenAI vs Agents: Deep Learning = the base technology (neural networks that learn from data). GenAI = using deep learning to create new stuff like text or images. Agents = AI that can plan, use tools, and take multiple actions on its own to complete a task.

[Aman Kumar Srivastava @botz_coder]

Tokenization + hallucination fixes: Tokenization = breaking text into small pieces (words/sub-words) so the model can read it. Hallucination fixes = giving the model real documents to read from (RAG), keeping answers less "creative" (low temperature), and double-checking facts after generation.
LLM usage in apps (chatbots): You take the raw LLM and add memory (so it remembers the chat) and instructions (system prompt) so it behaves like a proper chatbot instead of just answering once.
LangChain (chains vs graphs): Chains = do step 1, then step 2, then step 3, in a fixed order. Graphs = the flow can branch, loop back, or skip steps based on what happens — more flexible, like a flowchart.
Prompt engineering: Carefully writing your instructions/examples so the AI gives you the answer you actually want.
Transformers + embeddings: Transformers are the AI architecture that looks at all words in a sentence together and figures out which words matter to which. Embeddings are just numbers (vectors) that represent the meaning of a word/sentence so similar things end up with similar numbers.
Temp vs top-k vs top-p: All three control how "random" or "safe" the AI's word choices are. Temperature = overall randomness dial (low = safe, high = random). Top-k = only pick from the top K most likely next words. Top-p = only pick from words that together make up P% probability (adapts automatically).
Chunking strategies: Before feeding documents to the AI, you cut them into smaller pieces so it's easier to search and fits in the AI's memory limit.
Activation functions (transformers): A math function inside the model that decides how much a neuron "fires." Transformers mostly use GELU (or similar) to add non-linearity so the model can learn complex patterns.
Gradient boosting: A method where you build many small decision trees one after another, each one fixing the mistakes of the last one, to get a really accurate final prediction.
MCP/tools integration: A standard way for an AI model to call outside tools (like search, calculator, database) so it's not limited to just its own knowledge.
DL vs GenAI vs Agents: Deep Learning = the base technology (neural networks that learn from data). GenAI = using deep learning to create new stuff like text or images. Agents = AI that can plan, use tools, and take multiple actions on its own to complete a task.

[Aman Kumar Srivastava @botz_coder]

@BrrandonHQ Thanks, not looking for that right now!

[Aman Kumar Srivastava @botz_coder]

Someone built a tool that turns any GitHub profile into a FIFA Ultimate Team card 🃏⚽
Rated out of 99, straight from your real GitHub activity:

Try this out: https://t.co/TNWLtODvt5

Commits → PAC
Stars → SHO
PRs + followers → PAS
Language diversity → DRI
Reviews + issues → DEF
Lifetime contributions → PHY

Tried it on myself — fun concept, cool execution. 1k+ stars on GitHub in just 2 days.

#buildinpublic #GitFut #OpenSource #IndieHackers #DevTools

[Aman Kumar Srivastava @botz_coder]

@awwmishaaa congrats @awwmishaaa for this milestone 🎉

[Aman Kumar Srivastava @botz_coder]

@mange_manali Thanks Manali

[Aman Kumar Srivastava @botz_coder]

@trying_to_exits Hi Abhishek 👋, let's connect

[Aman Kumar Srivastava @botz_coder]

@coffee_0708 check this out : https://t.co/6W7JMoAdcK

[Aman Kumar Srivastava @botz_coder]

@vibsbuild Check out this one
https://t.co/lsjTzVuuhr