Sukhjit Singh

LLM & ML Engineer with 2+ years of production experience across NLP, computer vision, and intelligent automation — built and shipped end-to-end systems at Genpact (global professional services) serving US and UK enterprise clients. Designed an 8-class imbalanced NLP classification pipeline processing 50,000+ emails/month, lifting accuracy from 65% (manual routing baseline) to 92%+ with Scikit-learn and FastAPI. Completed that engagement in April 2026; pursuing high-impact ML/LLM engineering roles at product-led companies.

Published researcher with 14 international publications including a Springer Nature peer-reviewed journal paper and dual full-paper presentations at IAC 2024, Milan. Contributor to ISRO's operational SAR target recognition pipeline.

ML Engineer - Genpact (Global Service) - Hybrid Remote-first, US/UK clients

(2024-07 - 2026-04)

Project delivery cycle completed April 2026; currently available and actively interviewing.

• Engineered a production 8-class NLP email classification system independently — designed the full architecture (Python, Scikit-learn, FastAPI), lifted accuracy from 65% rule-based baseline to 92%+ on an imbalanced dataset of 50,000+ emails/month, reducing misrouted tickets by ~30% and eliminating 10 hrs/week of senior analyst review time.
• Consolidated 5+ fragmented HR data sources into scalable ETL pipelines (Python, SQL, REST APIs, Power Automate) — automated 120 hrs/month of processing, streamlining reconciliation errors by 40%, and established a single source of truth for workforce analytics.
• Deployed Power BI and Tableau dashboards on automated refresh cycles — compressed report generation from 3 days to same-day delivery, enabling real-time strategic decisions for senior leadership.
• Containerised 3 production ML services with Docker and integrated MLflow experiment tracking — cut model redeployment time from 3 days to under 4 hours by standardising environment configs and versioning artifacts across the full model lifecycle.
• Orchestrated LLM-powered automation pipelines (Claude API, n8n, Power Automate) for end-to-end workflow management — eliminated manual intervention touchpoints across cross-functional operations for US and UK client teams.

Research Intern - Indian Space Research Organisation - Hybrid

(2023-09 - 2024-02)

• Resolved critical out-of-distribution failure in SAR Automatic Target Recognition by introducing DCGAN-based synthetic augmentation on a 10,000+ image dataset — lifted accuracy from 81% to 84% and improved out-of-distribution detection by 7% as the sole student contributor on a team of ISRO scientists.
• Accelerated model iteration by building an end-to-end training pipeline covering data preprocessing, GAN augmentation, hyperparameter search, and evaluation (F1, ROC-AUC) — shortened per-experiment time by 35%, enabling 3× more ablation cycles within the same timeframe.
• Trimmed false positives by 12% in operational defence target recognition using ResNet-inspired CNN classifiers with skip connections across 4 scenarios — findings adopted into IIST's operational protocols.

Research Trainee – Statistical & ML Modelling - IISA, DRDO - On-site

(2023-05 - 2023-06)

• Simulated 6 military-operation scenarios (SciPy, Scikit-learn) across 7 environments and built a Python sensitivity-analysis framework — identified top 4 parameters driving simulation variance, optimising re-simulation cycles by 18%, and shortened feedback loop from 2 days to under 6 hours. Results validated by senior DRDO researchers.

M.Sc. - Applied Mathematics (Hons) - University of Delhi (2022-08 - 2024-05)

B.Sc. - Mathematics (Hons) - University of Delhi (2018-07 - 2021-06)