Experiments in
AI & learning.

I'm Sriram — I build things to understand them. This is my scratchpad: half-finished ideas, things I'm poking at, notes I'd otherwise forget.

LLMs agents RAG fine-tuning evals product

experiments // what I'm poking at

CMCSA 10K — Comcast Strategic Intelligence

RAG pipeline over 24 years of Comcast annual filings (2001–2024). Ask natural-language questions and get GPT-4o answers with source years cited. Embeddings stored in ChromaDB; retrieval via LlamaIndex.

Python FastAPI LlamaIndex ChromaDB GPT-4o text-embedding-3-large

active

Experiment name goes here

A sentence or two about what this is and what I learned. Real content coming soon.

tool tool

paused

Experiment name goes here

A sentence or two about what this is and what I shipped. Real content coming soon.

tool tool

done

CMCSA 10K // comcast strategic intelligence

This demo queries a RAG pipeline built over 24 years of Comcast (CMCSA) 10-K SEC filings. Each filing was chunked, embedded, and indexed so you can ask strategic questions and receive grounded answers with source years cited.

Questions are answered by GPT-4o, which synthesises the most relevant retrieved passages and produces a concise response. The backend runs as a FastAPI service; retrieval is handled by LlamaIndex over a ChromaDB vector store with text-embedding-3-large embeddings.

GPT-4o text-embedding-3-large LlamaIndex ChromaDB FastAPI Python OpenAI API

Ask anything about Comcast's annual filings.
e.g. "How has broadband revenue trended since 2010?"

notes // things I keep relearning

—

Note title. A quick observation or thing I figured out that was worth writing down. Real notes coming soon.

—

Note title. A quick observation or thing I figured out. Real notes coming soon.

—

Note title. A quick observation or thing I figured out. Real notes coming soon.

stack // what I'm using lately

model

—

coming soon

framework

—

coming soon

infra

—

coming soon

tooling

—

coming soon