Looking for a job

Anuj Agrawal

@anujagrawal

Message

Data analyst at FRND who ran controlled tests across 1.5M users, lifting games played 326.5% and coin spend 259.6%.

India

Message

At FRND, I engineered controlled ABBA experiments across 1.5M users to test more than 150 creatives, lifting games played by 326.5% and coin spend by 259.6%. I also built ClickHouse SQL workflows across 74M+ rows and cohort analyses to track downstream impact beyond CTR.

At Hyperleap AI, I built an IIT-JEE AI tutor for physics, chemistry, and mathematics using open-source LLaMA and Mistral models. I developed a retrieval-augmented generation pipeline over a curated 1,000+ question knowledge base and benchmarked the live tutor against ChatGPT on 50+ problems.

In my research internship with Professor Philip Treleaven at University College London, I designed a declarative architecture for privacy-preserving LLM fine-tuning and co-authored SSRN-published research. I’ve also built a full-stack event management platform and worked on deep-learning systems for sound analysis and pronunciation scoring.

Experience

Work history, roles, and key accomplishments

Research Intern

University College London

Mar 2026 - Aug 2026 (5 months)

Designed a dual control-plane architecture for secure data-local LLM fine-tuning under regulatory and sovereignty constraints. Integrated DP-SGD, Federated Learning, LoRA/QLoRA, and Federated Computing as Code into a declarative architecture for distributed privacy-preserving model training.

LLM Fine tuning Differential Privacy Federated Learning LoRAs QLoRA Secure Aggregation Governance as Code

Data Analyst Intern

FRND

May 2026 - Jul 2026 (2 months)

Engineered rank-matched ABBA experiments across 1.5M users, testing 150+ creatives, driving +326.5% games played and +259.6% coin-spend lift. Built production ClickHouse SQL workflows on 74M+ rows and performed RCA & cohort analysis to improve D1 retention from 8-9% to 14-15%.

ABBA Experiments Clickhouse SQL Google Apps Script JavaScript Data Analysis Cohort Analysis Retention Analysis

Machine Learning Intern

Hyperleap AI

May 2025 - Jul 2025 (2 months)

Built an IIT-JEE specialized AI tutor for PCM, integrating open-source LLaMA & Mistral LLMs to generate structured, step-by-step solutions via an interactive chatbot-style interface. Engineered a RAG pipeline using SBERT embeddings and ChromaDB to semantically retrieve relevant problems from a curated 1,000+ question JEE knowledge base.

LLaMA Mistral RAG SBERT ChromaDB Prompt Engineering Python LLM