hi, i am sajal and this is the place that'll probably have all of my digital footprint? ugh okay my twitter has that. okay so maybe this is the place where i can dump my raw thoughts? got substack for that as well. umm maybe my projects? github got that. movies? well i am already very irregular at logging on letterboxd. so idk why am i creating this personal website, and it's already having an existential crisis. well i've always wanted to create one and it looks fun and having fun is legit a good use-case (a recent discovery) so here it is. why are you still here, reading this, this is the most boring section probably, go on to some other section, i am out of thoughts here.

what i do

ongrid (founding ai engineer, jan 2024 – present)

  • built the document intelligence system that now processes 10L+ requests/month across 15+ doc types.
  • bgv initiation portal: all document data extraction was manual before this. now it's fully automated. 300% throughput increase.
  • built an impersonation detector that silently joins interviews, runs face-match in real time, and tracks candidates across rounds. creepy in the best way.
  • took our chatbots from basic RAG to full agentic workflows. MCP, tool calling, SSE streaming, the whole stack.

if you're working on something hard and interesting and need someone who's obsessed, my email is right up there.

apart from work - movies, f1, gym, cocktails, and shitposting on my alt.

projects

llama2 from scratch first principles learnings is fun actually
transformer from scratch all hail karpathy!
cornelia a side project that gave me my biggest learnings/month ratio
llmd because feeding raw HTML to a model is messy, and i am lazy
healthcare chatbot handles scanned PDFs, insurance policies, prescriptions
blackbox labs i grew my twitter from 0 to 10k in 5 months, so i kinda knew what worked and what not, so automated all of my wisedom with this.
jobjesus built in a hackathon for fun, after OpenAI dropped API access

writings

breaking bad on fine-tuning an llm
google vision ocr vs open source ocrs
prompt engineering might be all you need
llm inference 101
finetuning llms — lora & qlora
quantization
ml
llms
llama from scratch

reading