Headline

AI Engineering Intern @Twigg | Ex - AI Research Intern@Superkalam(YC W23) | SIH 2023 Finalist |

About

Quick background: - IIT Madras BS student (final year) - AI Research Intern SuperKalam (YC 23): AI mentor for Upsc students - AI Research Intern (USC Research): Published work on LLM context alignment (https://arxiv.org/abs/2507.02949) - AI/ML Intern Torchit: Worked on Multimodal AI applications for visually impaired; project got UN award - SIH Finalist (ISRO PS): LLM fine-tuning and NER - Amazon ML Challenge: Top 82/10k (OCR + multimodal LLMs) - Hackathons: 3rd at Pixelmind, Top 17 in GST hackathon, Google Regional qualified, 5+ finals

Location

Gorakhpur

AI stack

aistudio

claude code

claude

hugging face

langchain

openai

Work experience

AI Engineering Intern at Twigg, Apr 2026 - Sep 2026 · 6 mos, Finance + AI

AI Research Intern at SuperKalam (YC W23), Dec 2024 - Apr 2026 · 1 yr 5 mos, Remote, Improved model answer generation and evaluation, implementing intelligent caching + multi-chain prompting, boosting rating from 4 → 4.5. Designed and optimized a RAG pipeline for educational queries (PYQs, structured outputs, retrieval vs generation decisioning). Enhanced decider chain logic for efficient retrieval, consistency, and latency optimization. Worked on MCQ quality evaluation pipelines, including benchmark creation and large-scale prompt-driven improvements. Contributed to Ethics and optional answer evaluation systems and in-copy handwritten assessment workflows.

AI Research Intern at University of South Carolina, Jul 2024 - Dec 2024 · 6 mos, Remote, Researching context alignment in large language models under Prof. Dr. Amitava Das and Dr. Amit Sheth, mentored by Aman Chadha, Vinija Jain, and Aishwarya Reganti. - Built RAG pipeline to analyze context alignment across multiple open-source and closed-source LLMs (Gemini, Claude, Llama, Mistral etc.) - Systematically analyzed divergence patterns in foundation models to understand how and why context-response misalignment occurs - We developed "RAGability" - a novel metric to quantify contextual divergence in LLM outputs Designed experimental pipelines to optimize context-response alignment, improving factual consistency

Research Intern at Indian Institute of Technology, Mandi, May 2024 - Dec 2024 · 8 mos, Mandi, Himachal Pradesh, India, Built end-to-end machine learning pipelines for brain-computer interfaces using EEG data. - Developed ML pipeline for 8-channel EEG devices achieving 87% accuracy in cognitive load detection through optimized signal processing and feature extraction - Collected and processed EEG data from 50+ students, implementing comprehensive feature engineering for distraction and mind-wandering detection - Conducted extensive literature review of 10+ research papers to identify optimal experimental methodologies for cognitive assessment

AI/ML Intern at TorchIt, Jan 2024 - May 2024 · 5 mos, Ahmedabad, Gujarat, India, Developed "Jyoti" with team - assistive vision technology for the visually impaired using multimodal LLMs. Project won UN's SGD Digital Pioneers Award. - Built multimodal AI systems combining open source multi modal language models to provide real-time scene understanding and navigation assistance - Collaborated with AI/ML team to develop educational platform with automated quiz generation and content summarization using OCR and NLP pipelines - Contributed to accessibility-focused AI solutions

Education

Bachelor of Science - BS, Data Science and Programming, Indian Institute of Technology Madras

10th, SSDT Inter College

Public links

github: https://github.com/Abhay-404