Himanshu Prashar

LLM evaluation and coding benchmark specialist with 2+ years of experience designing, implementing, and reviewing repository-level software-engineering and machine-learning evaluation tasks for AI agents. Authored and validated 50+ containerized benchmark tasks across Python, Go, Rust, C/C++, Ruby, and ML evaluation workflows. Experienced in adversarial task design, hidden-test development, rubric design, Docker-based reproducibility, automated verification, oracle validation, Harbor evaluation runs, deterministic scoring, and model failure analysis.

Also brings hands-on AI research experience in medical image segmentation, computer vision, biomedical signal processing, and time-series modeling through research at CURAJ, NIT Jamshedpur, IIT Kharagpur, and IIITDM Kurnool.

LLM Coding Benchmark Contributor & Quality Reviewer at Multiple AI Evaluation Platforms, including Snorkel AI Terminal-Bench (2024-01 – Present)

• Authored and validated 50+ containerized, repository-level coding tasks used to evaluate and stress-test frontier LLM agents across Python, Go, Rust, C/C++, and Ruby.
• Designed hard but solvable tasks involving coupled defects, stateful behavior, concurrency, scheduling, binary formats, parsers, compiler tooling, scientific computing, and distributed-system workflows.
• Developed public and hidden test suites, executable rubrics, deterministic verifiers, and fail-to-pass checks that distinguish complete implementations from superficial patches.
• Performed instruction-sufficiency, solvability, anti-leakage, originality, and difficulty reviews; diagnosed flaky tests, environment defects, unintended shortcuts, and recurring agent failure modes.
• Built reproducible Docker environments and validated tasks through clean-container, offline, baseline, oracle, and repeated end-to-end Harbor runs.
• Reviewed GitHub repositories, commits, diffs, and CI results to assess implementation quality and provide actionable technical feedback.

Research Intern – Oral Cancer Histopathological Image Analysis at National Institute of Technology Jamshedpur (2025-12 – Present)

• Developing methods for automated detection and segmentation of oral cancer from whole-slide histopathology images using pretrained CNN and vision-transformer backbones.
• Addressing stain variation, high-resolution tiling, tissue-level segmentation, and class imbalance through preprocessing and augmentation pipelines.

Research Intern – EEG Signal Analysis for Autism Detection at Indian Institute of Technology Kharagpur (2025-08 – 2025-11)

• Investigated P300 responses as neurological markers for autism using 8-channel EEG recordings from 15 subjects across multiple sessions.
• Designed a lightweight CNN–Transformer model for spatial–temporal feature learning, achieving 96.64% test accuracy with subject-wise, session-wise, and cross-validation evaluation.

Deep Learning Research Intern – Human Activity Recognition at Indian Institute of Information Technology, Design and Manufacturing Kurnool (2025-05 – 2025-07)

• Proposed a WaveNet–BiLSTM–GRU architecture for wearable-sensor activity recognition and built an end-to-end accelerometer and gyroscope processing pipeline.
• Evaluated the model on UCI HAR, WISDM, and MHEALTH using confusion matrices, ROC curves, and per-class reports; manuscript prepared for submission.

Integrated M.Sc. in Computer Science – Central University of Rajasthan (CURAJ) (2021-01 – 2026-01)

Higher Secondary in CBSE – Jawahar Navodaya Vidyalaya (2020-01 – 2020-12)