Shivank Singh

AI Full Stack Developer with hands-on LLM post-training experience: RLHF-style evaluation, SFT trajectory creation, and rubric-based quality review, combined with production full-stack engineering (React, Node.js, MongoDB).

LLM Post-Training Intern - Ethara AI - Gurugram, India

(2026-01 - 2026-07)

• Evaluated LLM-generated responses across single-turn and multi-turn conversations against structured rubrics for RLHF preference-data pipelines, assessing contextual accuracy, coherence, and response consistency.
• Designed and refined evaluation rubrics using an atomicity-based, six-tier weighting framework to standardize scoring across multiple annotators and reduce inter-rater disagreement.
• Authored SFT (Supervised Fine-Tuning) training trajectories for an agentic AI system, writing multi-step task demonstrations used to fine-tune model behavior on real-world workflows.
• Performed side-by-side (SxS) comparative evaluations of model output variants and supported RLVR-oriented reward-signal validation to catch quality regressions before updates shipped.
• Conducted human-in-the-loop audits with domain-based taxonomy tagging to flag hallucinations, bias, and logical inconsistencies, documenting recurring failure patterns that fed back into alignment workflows.

Bachelor of Technology - Computer Science and Engineering - J.B. Knowledge Institute of Technology & Management (MD University) (2022-08 - 2026-06)