sess_042 · live model: pulkit-v2.6 loss ↓ 0.0042

> initializing agent… Pulkit Mangal

I build models that reason, pipelines that ship, and agents that get out of the way. Currently Data Scientist & SME at Scaler, finishing my BS in Data Science at IIT Madras.

view projects
attention weights
02

the .readme

I'm a data scientist who likes the boring parts — the data validation, the eval harness, the loss curves that refuse to go down. That's where the real work is.

Right now I'm at Scaler, writing scripts for AI engineering coursework and prototyping RAG pipelines and agents that actually fit into team workflows. Before that I was at Uncia (IITM Research Park) shipping anomaly detection and time-series forecasting to production.

I teach as much as I build — TAing Intro to Deep Learning, GenAI, ML Techniques, and Math II at IIT Madras. Translating a gradient-descent update into a sentence a first-year actually understands is its own craft.

  • / end-to-end ML pipelines, from notebook → production
  • / transformers, PEFT/LoRA, RAG, evals
  • / agents — tool-use, planning, guardrails
  • / data storytelling that holds up in a stakeholder review
03

the stack

organized by the projects that put it into practice
04

training log

05

model cards

06

/now

writing

scripts + lesson plans for Scaler AI Engineering

building

internal agents + RAG tooling — routing, retrieval, eval harness

learning

MLOps playbooks · LLM fine-tuning · long-context retrieval

location

Bangalore, IN · from Rajasthan · remote-flexible

07

pretraining

2022 — 2026

BS, Data Science & Applications

Indian Institute of Technology Madras

in progress
08

open a channel

pulkit@portfolio ~/inbox $

let's build something
that actually ships.