Govind Arun Nampoothiri
MSc Data Science graduate building applied AI and data systems that connect data pipelines, modeling, and decision-support workflows in product settings. Experienced across end-to-end product delivery, from data modeling and analytics to deployment, experimentation, and system design.
Interested in Data Scientist, Machine Learning Engineer, or AI Engineer roles where strong product thinking and technical ownership matter.
Experience
AI Engineer Intern
The Product Works
Chennai, India
Architected an end-to-end move-telemetry and analysis foundation for Chessy, persisting per-move evaluations, aggregating game-level diagnostics, and feeding per-student coaching dashboards with causal signals, ratings, blunder patterns, and practice positions derived from student games.
Designed a hybrid inference deployment combining client-side WASM with server process pools and provider-agnostic LLM routing, with prompt caching to lower token cost, optimised for sub-500 ms feedback on tablets, enabling high-quality, real-time coaching for young learners.
Led modular integration architecture using versioned contracts and shared schemas, decoupling frontend portals from core game logic to enable rapid, independent cross-team delivery.
Engineered production infrastructure (Supabase, FastAPI, Redis, CI/CD, MLOps practices) scaled and cost-optimised for 10,000+ concurrent students while providing reliable monitoring and low-latency APIs.
Shipped an agentic AI assistant with a provider-agnostic LLM layer, streaming chat API, dynamic system prompt assembly with context pruning and URL-based pronoun resolution, multi-step tool orchestration, role-based tool filtering, safety probing with 3-miss aborts, and PII-safe architecture groundwork.
Built the telemetry data model and Chess Base-style coach knowledge base to store positions, move trees, motifs, annotations, and curated training sets used for targeted search, analysis, and coach-driven drill generation.
Data Science Intern
Fittlyf
Chennai, India
Engineered an end-to-end A/B/Multivariate testing platform (Streamlit, Sci Py), automating statistical analysis and sequential testing to reduce decision time by 40% and false-positive rates by 25% across 1,000+ experiments.
Developed organization-wide SRM detection dashboards (Tableau, SQL) to automate sampling health monitoring, reducing bias investigation time by 10% while standardizing experiment workflows and documentation.
Education
The University of Edinburgh
MSc
Grade: Merit
Dissertation: Causal Inverse Reinforcement Learning for Robust Reward Recovery
Christ University
BSc
First Class with Distinction, CGPA: 3.8/4.0
Skills
Projects
Real-Time Payment Fraud Detection System
Built a scalable fraud detection pipeline using Graph Neural Networks and Reinforcement Learning, handling 590 k+ transaction nodes with <28 ms P 95 inference latency on a single T 4 GPU, matching real-time deployment latency constraints and achieving a 22% improvement in G-Means (57%) over baseline on highly imbalanced (28:1) data through Adaptive Majority Downsampling and an autonomous RL agent.
Deployed an API using FastAPI, Docker, and Redis with MLflow model versioning, automated CI/CD (GitHub Actions), and feature attribution via GNNExplainer for every flagged transaction to support regulatory compliance.
Customer Journey Intelligence: Graph-RAG System
Demo
Developed a Graph-RAG retrieval pipeline (FAISS, Sentence-Transformers, NetworkX) that outperformed standard vector RAG on temporal queries by extracting path-aware journey patterns (churn sequences, cross-sell paths) from 85 k+ clickstream events across 5 k users.
Powered 5+ cohort-comparison quantitative analytics with cohort comparison retrieval using Groq/Llama 3 8 B, generating segment-specific insights (conversion rates, journey frequencies) instead of generic summaries, deployed on Hugging Face Spaces with a FastAPI backend, Streamlit frontend, Docker containerization, CI/CD via GitHub Actions, and unit tests validating retrieval logic.
Banking Customer Retention: Causal Targeting Engine
Demo
Designed a causal inference framework (T-Learner with XGBoost) for precision marketing, achieving 3% global lift and 86% reduction in marketing waste by isolating a 14% “persuadable” customer segment from neutrals and hard negatives.
Productionized the MLOps pipeline with FastAPI, featuring interactive Qini-curve dashboards, for real- time inference and Docker for deployment, validating model stability and interpretability via Do Why refutation tests and SHAP analysis.