Learning new ideas and fiddling with them

When you visit a site what happens ?

*The html , js , css gets transferred to the browser using http request and then the chrome engine runs that and display the content

How does image upscaling works?

How does real time stt works ?

We store the 500ms chunk in /tmp file , send it to transcribe and then repeat this until we close it .. or a VAD (voice activity detection) is used to check till when the voice is detected ..

Network differences between a sse and stdio protocol ?

SSE : Its a long lived HTTP GET request , that the client sends to the server and server sends the data back to it ..

STDIO : Its a local IPC ( inter process communication ) , Standard Input/Output , its a way for programs to read and write data , typically in command line or process-based environment ,

stdin : input stream ( keyboard , input mic ) stdout : output stream ( printed output , transcription result ) stderr : error messages

Its for internal communication , lets suppose a server (whisper model) -> audio chunk -> subprocess transcription -> transcribed output -> send to frontend using SSE or websocket

Now lets see an LLM request:

1. /post request from frontend
2. sends that to LLM
3. receive chunks -> send back to frontend via SSE

Hosted LLM sending chunks : Recieves prompt -> predicts tokens clubs and sends it back -> (sse) here the client is the backend that called that LLM ..

Local LLM sending chunks : Recieves prompt (stdin) -> predicts tokens -> sends chunks (stdout) -> stdio operation

# Start local LLM binary
llm_process = subprocess.Popen(
    ['./llama.cpp', '--model', 'mistral.gguf'],
    stdin=subprocess.PIPE,
    stdout=subprocess.PIPE,
    text=True
)

# Send prompt
llm_process.stdin.write("Tell me a joke\n")
llm_process.stdin.flush()

# Read output line by line (like tokens)
for line in llm_process.stdout:
    yield f"data: {line}\n\n"  # <-- SSE back to frontend