Alok Kushwaha

I’m an NLP and ML engineer focused on taking a modelling problem from raw data to a working system—training code, APIs, and a frontend people can open in a browser. My core strength is owning features end-to-end and shipping things that perform in real workflows.

I’ve published a peer-reviewed paper on financial contract risk analysis using Transformer-based clause modelling and graph neural networks. Alongside that, I’ve built and deployed four projects spanning financial risk modelling, legal NLP, and multi-agent AI.

In internships, I improved downstream model quality by building LLM fine-tuning preprocessing pipelines, and I benchmarked RAG workflows across FAISS, Chroma, and Pinecone to optimize retrieval precision/recall. I also annotated multi-million-record datasets with inter-rater reliability scoring to surface systematic labelling errors and reduce ingestion risk.

From an engineering perspective, I’ve owned ingestion pipelines, scheduling, and error logging, cutting manual preparation time by ~8 hrs/week. I’ve also tuned data and APIs for performance—diagnosing slow SQL, redesigning indexing, and reducing latency by 30%—while delivering full pipelines like OCR → clause segmentation → graph dependency modelling → FastAPI + Next.js.