Backend Engineer · Python · Django · PostgreSQL · Docker
LLM Evaluation · golden sets · LLM-as-judge · eval harnesses
_{CS undergrad at Scaler School of Technology & BITS Pilani · Bangalore, India}

I build backend systems that actually ship. As a solo freelancer I took a B2B e-commerce platform from zero to production in 6 weeks — Django REST API, React frontend, PostgreSQL — now live at pronounjeans.com and serving 50+ active B2B users. Most of my work sits at the intersection of clean API design, concurrency correctness, and containerized deployment.

I also build evaluation for LLM systems — hand-labelled golden sets, blind LLM-as-judge scoring, and judge-vs-human agreement studies, because a system you can't measure isn't a system you can trust.

Currently: building CodeGraph — a static-analysis engine that maps code relationships using Python's AST, with Celery + Redis handling heavy parsing off the request thread.

Open to: backend / full-stack internships at early-stage startups where I can own features end-to-end, and contract work on AI evaluation and benchmark design.

Freelance · Live in production · Mar–Apr 2026

  🌐 Live Site  ·  💻 Code

Designed, built, and deployed the entire platform solo, end-to-end.

• 50+ active B2B users on the live platform today
• 4 modular DRF apps — product catalog, cart, checkout, order management — exposing REST APIs to a React frontend
• 3-service deployment pipeline owned end-to-end: Django on Render Pro, React on Vercel, PostgreSQL on Supabase
• Diagnosed and fixed a production static-file outage via Django's collectstatic pipeline + Whitenoise — eliminated 100% of frontend asset errors

_{Django Django REST Framework PostgreSQL Supabase React Whitenoise Render Vercel}

Sept 2026

  💻 Code

An AI support agent built on the Customer Support on Twitter dataset — and, more to the point, the evidence for whether it can be trusted.

• 10-intent taxonomy derived from the data, not assumed up front
• Hand-labelled golden set with a written codebook, plus human reply scores
• Blind LLM-as-judge harness, validated against human raters for agreement
• Documented failure analysis — including that the agent fabricated employee initials in 91% of replies — and a written account of why the headline metric misleads

_{Python LLM-as-judge golden sets inter-rater agreement failure analysis Make}

In progress · June 2026 – Present

  💻 Code

Code-parsing system built on Python's ast module to extract structural relationships from source files.

• Async parsing via Celery + Redis — resource-intensive jobs never block the main application thread
• Multi-container stack — frontend, backend, and Celery workers containerized and orchestrated with Docker Compose
• RESTful API serves parsed AST data and code metrics to a React + Vite frontend that renders the graphs

_{Python Celery Redis Docker Docker Compose React Vite}

Dec 2025 – Feb 2026

  💻 Code

Full-stack banking application — designed and built the React frontend and Django REST backend independently.

• Atomic fund transfers using Django ORM's select_for_update() — prevents double-spend under concurrent requests
• PostgreSQL schemas with integrity constraints enforcing valid financial states at the database level
• Structured into modular Django apps for clean separation of concerns

_{Django PostgreSQL REST APIs Docker React}

Languages   Python JavaScript SQL Java

Backend   Django Django REST Framework Celery

Frontend   React Vite

ML & Evaluation   pandas Jupyter DuckDB golden-set design LLM-as-judge inter-rater agreement

Data & Infra   PostgreSQL Redis Supabase Docker Docker Compose Linux Git Postman

Focus areas   REST API design · async task processing · concurrency control · database schema design · containerization · evaluation design

_{Reach me at talindaga692@gmail.com — I reply fast.}