Hey, I'm Aman Kushwaha
Results-driven Data Science undergraduate with hands-on experience in machine learning, AI systems, full-stack development, and open-source projects. Passionate about building scalable solutions, learning emerging technologies, and collaborating with diverse teams to solve complex challenges. Seeking opportunities to contribute technical expertise while continuing to grow as an engineer.
Terminal
VoxRAG
VoxRAG is a voice- and text-enabled Retrieval-Augmented Generation system built on MSMARCO-XI. Speak or type in English, हिन्दी, मराठी, বাংলা, മലയാളം, ગુજરાતી, অসমীয়া — the pipeline transcribes, retrieves grounded passages from ~3.2M (32 lakh) embedded records, and streams a cited answer back, targeting <200 ms retrieval / time-to-first-token.
AAIPL Tournament: Dual-Agent Fine-Tuning Solution
Won the AAIPL Hackathon (1st place out of 120 teams) by building a dual-agent LLM system using Qwen2.5-14B-Instruct. Developed end-to-end fine-tuning pipelines with LoRA and 4-bit quantization to train a Question Agent for generating MCQs and an Answer Agent for solving them with structured reasoning. Implemented robust data cleaning, custom dataset formatting, and efficient training workflows for high performance under constrained compute.
Genome AI: Cattle Breed Classification from SNP Data
A reproducible machine learning pipeline for cattle breed classification from SNP genotype data, featuring CNN and Transformer baselines. Includes data validation, preprocessing, training, and benchmarking workflows to compare local vs long-range genomic pattern modeling.
Conveyor Belt Detection
Labeled custom conveyor belt data and trained a YOLOv8 model from scratch. Increased detection accuracy from 72% to 80% with model tuning and augmentation.