Headline

Fraud & Risk Analytics | Production ML for anomaly detection | Senior Development Analyst @ Accenture | ISB AMPBA

About

I build systems that catch things that shouldn't be happening. For the last five years that has meant fraud and anomaly detection on high-volume transaction data supervised models, Isolation Forest, and NLP pipelines running in production against 50M+ records and 20,000+ events a day. What I've found is that the hard part is rarely the model. It's everything around it: which threshold you pick, how many false positives your analysts can actually absorb, when the model has drifted far enough to need retraining, and whether anyone can explain a flagged decision six months later in an audit. Some of what that's looked like in practice: - Supervised + Isolation Forest models across multi-asset workflows — 15% reduction in fraud-related losses, with threshold governance to keep it there - A real-time detection model on 50M+ records at 91% precision / 85% recall, under 2 second latency - An NLP pipeline (spaCy, Hugging Face) pulling risk signals out of unstructured communications 60% less manual review effort - A BigQuery pipeline standardising 20,000+ daily events into model-ready risk vectors I'm currently doing the AMPBA at ISB alongside work. Interested in fraud risk, credit risk, and applied ML roles at product companies where risk decisioning is core to what gets built. anmoljindal3977@gmail.com

Location

Gurugram

AI stack

claude code

claude

codex

langgraph

langsmith

openai

Work experience

Senior Analyst at Accenture, Sep 2025 - Present · 1 yr, Gurugram, Fraud & Risk Detection Platform I own the anomaly detection roadmap and the models that sit on it — from prioritisation through production monitoring. - Architected and operationalised supervised + Isolation Forest models across multi-asset workflows over three quarters. 15% reduction in fraud-related losses, with ongoing threshold governance so it holds. - Run production detection at steady state: alert volume across analyst queues, threshold recalibration driven by analyst feedback, monthly review cadences for sustained accuracy and audit readiness. - Built an NLP pipeline (spaCy, Hugging Face) extracting risk signals from unstructured communications, with explainability and threshold-tuning controls. Cut manual review effort 60% and set up monthly retraining. - Built a BigQuery pipeline standardising 20,000+ daily events into model-ready risk vectors 20% improvement in data reliability across two production cycles. - Prioritised the roadmap using RICE and MoSCoW against quarterly objectives; authored the PRDs and drove alignment across Engineering, Data Science, DevOps and Platform Ops in a 6-person pod spanning US, EU and India. Python · SQL · BigQuery · Scikit-Learn · spaCy · Hugging Face

Application Developer at Accenture, Jan 2024 - Aug 2025 · 1 yr 8 mos, Gurugram, Haryana, India, -Translated trading, risk, and compliance requirements into backend components within a production trading and risk-management platform. -Built and optimized scalable data pipelines using Python, BigQuery, and cloud services to process high-volume trade data. -Engineered feature transformations to support fraud detection and anomaly-detection models, enabling downstream analytics and alerts. -Integrated ML outputs into Python-based analytics services and dashboards, ensuring reliable data delivery across systems. -Supported early experimentation with LLM-assisted analysis for summarizing trade patterns and improving investigation workflows.

Consultant at Wipro, Oct 2022 - Dec 2023 · 1 yr 3 mos, Gurugram, Real-Time Trade Integrity & Market Risk - Built an Isolation Forest-based real-time anomaly detection model on 50M+ records. Tuned to 91% precision at 85% recall with sub-2-second detection latency, which was the actual constraint a model that's right but late is useless in a live system. Stabilised portfolio volatility 18%; established quarterly retraining. - Architected regulatory data validation and duplicate detection frameworks over three quarters, adapting ingestion rules and alert thresholds to changing compliance requirements. Automated manual reconciliation, reduced audit escalations. - Diagnosed transaction failure patterns to find execution gaps and liquidity risk, then redesigned the settlement workflow around what the analysis showed — 20% improvement in auto-settlement success rates. - Migrated legacy batch pipelines into modular Azure Data Factory and Logic Apps workflows with CI/CD, standardising schema validation across disparate vendor feeds. Python · PySpark · SQL · Azure Data Factory · Oracle

Project Engineer at Wipro, Jul 2021 - Oct 2022 · 1 yr 4 mos, Gurugram, Haryana, India, - Setup an automated system using Python for managing trader access to trading management tool based on their access on trading floor. - Developed and optimised data processing pipelines by adding validation checks to data quality workflows. Hence, improving execution time by 40% and preventing NULL/Sparse data from flowing. - Partnered with Data Engineering, DevOps and analytics teams to ensure production stability, testing strategies and SLA adherence.

Software Engineer Intern at FrugalTesting, Jan 2021 - Jun 2021 · 6 mos

Data Science Intern at Flipr Innovation Labs, May 2020 - Jun 2020 · 2 mos

Education

AMPBA, Business Analytics, AI, GenAI, Indian School of Business

BTech - Bachelor of Technology, Computer Engineering, Thapar Institute of Engineering & Technology

Public links

github: https://github.com/anmol0709