AI/ML engineer Chicago, Illinois

I build AI systems that make complex work simpler.

I’m Sabarish Raja Ramesh Raja. I turn messy data and emerging AI capabilities into products people can actually use, from natural-language analytics to agent workflows and model evaluation.

Explore my work View résumé

LinkedIn ↗ GitHub ↗ Email ↗

Production AILLM applications and safe analytics

Independent builderMobile products and developer tools

Published researchTwo peer-reviewed ML papers

01 / Experience

From data to decisions.

SubjectToClimate

Remote · 2025–2026

AI & Automation Intern Feb–May 2026

Built a generative AI analytics platform that brought GA4, Search Console, FullStory, and Meilisearch data into Supabase/PostgreSQL. Nontechnical teammates could ask questions in plain English; Claude generated queries using the live schema, and a read-only database role protected production data.

Automated daily data syncs and built a climate-news workflow that scores stories by relevance, publishes qualifying content, and routes uncertain cases to editors.

SEO & Data Intern Oct–Dec 2025

Built a KPI dashboard covering 20+ metrics across eight state hubs and analyzed 5,000+ site-search queries to guide content decisions.

SubjectToClimate profile

Bytecompass

Feb–May 2024

Web Development Intern

Developed responsive React interfaces and integrated REST APIs for a user-management dashboard, including live profile updates and mobile layouts.

02 / Selected work

A few things I’ve built.

Production work, independent products, and the engineering behind reliable AI.

Applied AI · Data platform

Natural-language analytics at SubjectToClimate

Unified four data sources and built a Claude-powered question interface so staff could explore site performance without writing SQL. Live schema awareness, structured output, and a SELECT-only database role made the workflow useful and constrained.

Python · FastAPI · Claude API · PostgreSQL · Supabase

Work profile

Independent product · Mobile AI

StyleSense

A React Native wardrobe app that turns a clothing photo into structured data, applies wardrobe and occasion rules in PostgreSQL, then asks Claude to rank eligible outfits.

React Native · Supabase · Claude Vision

View StyleSense

Open source · Developer tooling

Claude-Hindsight

A local CLI that indexes coding-agent sessions into SQLite and surfaces a concise project briefing, dashboard, and instruction audit.

TypeScript · SQLite · CLI

Explore the code

Research engineering · Model evaluation

Adaptive Reasoning Budget

A controlled evaluation system for studying accuracy and token-use tradeoffs in small reasoning models on an 8 GB GPU. I built the budget controls, paired comparisons, and audit checks that distinguish valid answers from truncated output. The research is ongoing.

Python · PyTorch · Hugging Face · pytest

Agent workflow · Video

Marketing-video pipeline

Eight specialized agents turn a product brief into a short video. Typed handoffs, human approval points, and visual QA keep generated assets and final renders aligned.

CrewAI · Pydantic · Gemini Vision · Remotion

Also built CanterburyGPT ↗, a GPT-style transformer from scratch, and Inkwell ↗, a letter-writing web app.

03 / About

Engineering judgment matters.

I like the part of AI engineering where a promising demo has to become a dependable product: clean data, sensible constraints, meaningful tests, and a clear result for the person using it.

My work spans LLM applications, data pipelines, multi-agent systems, and model evaluation. I hold an M.S. in Computer Science from Illinois Institute of Technology and a B.Tech in Computer Science from SRM Institute of Science and Technology.

What I work with

AI & ML: Claude, Gemini, PyTorch, Hugging Face, model evaluation
Data & backend: Python, SQL, FastAPI, PostgreSQL, Supabase, ETL
Products & workflows: TypeScript, React Native, CrewAI, n8n, GitHub Actions

04 / Research

Published work.

First author · Springer · 2024

Analyzing the Computational Efficiency of LLM Models for NLP Tweet Classification During Emergency-Crisis

Compared transformer models for classifying disaster tweets, considering both predictive quality and compute.

Read paper

Co-author · AIP Conference Proceedings · 2025

A Scalable Ensemble Learning with Performance Evaluation for Credit Card Fraud Detection

Studied rare-event fraud detection with Isolation Forest, XGBoost, SVM, and imbalance-aware resampling.

Read paper

05 / Contact

Let’s build something useful.

I’m interested in AI/ML engineering and applied AI roles where reliable systems meet real business problems.

sabarishds03@gmail.com

LinkedIn ↗GitHub ↗Résumé ↗