Saksham Kumar Sharma

Machine Learning Engineer

Halethorpe, Maryland, United States4 yrs experience

AI EnabledAI ML Practitioner

Key Highlights

Expert in building scalable ML and data engineering solutions.
Proven track record in fraud detection and compliance automation.
Strong background in NLP and LLM workflows.

Stackforce AI infers this person is a Fintech-focused AI/ML Engineer with expertise in data-driven solutions.

Contact

Skills

Core Skills

Machine LearningData EngineeringNatural Language Processing (nlp)Research

Other Skills

PythonSQLPostgreSQLDeBERTa-V3LongformerDockerAWS LambdaLangChainFAISSStreamlitPlotlyHPCADA serversETL pipelinesTableau

About

Hi, I’m Saksham Kumar Sharma, an aspiring AI/ML Engineer with hands-on experience building machine learning and data-driven solutions end-to-end. I enjoy taking real-world problems from raw data to deployed insights: building reliable ETL pipelines, engineering features, training and evaluating models, and iterating through error analysis to improve performance. I’m especially interested in applied ML and NLP/LLM workflows where strong engineering meets measurable impact. What I bring: ★ Machine Learning: classification/regression, tree-based models & ensembles, model selection/tuning, and metric-driven evaluation (ROC-AUC, precision/recall, MAE/RMSE). ★ NLP & LLM workflows: prompt strategies, dataset-driven evaluation, and building structured pipelines for language understanding and reasoning tasks. ★ Data Engineering: ETL pipeline development, SQL analytics, and large-scale data processing using Python and tools like Spark/Hadoop. ★ Engineering mindset: clean, reproducible experiments, debugging, automation, and building systems that are reliable and scalable. I also enjoy teaching and mentoring—breaking down complex ideas into simple steps and supporting others in building strong problem-solving fundamentals. I’m open to opportunities in AI/ML Engineering, Applied ML, and NLP/LLM engineering. Outside of work, I enjoy exploring new places, cooking, and playing table tennis.

Experience

4 yrs

Total Experience

1 yr 3 mos

Average Tenure

Current Experience

Kpmg us

AI/ML Engineer

Jan 2025 – Feb 2026 · 1 yr 1 mo

Developed an AI-driven platform analyzing transactional and financial statements to detect fraud patterns, automate compliance checks, and
enhance audit efficiency, collaborating closely with risk, compliance, and data governance teams.
Built scalable data ingestion pipelines using Python (PyMuPDF, pandas) and SQL, processing 10M+ DOCX, PDF, and HTML financial records, designing
relational schemas in PostgreSQL for structured storage and traceable pipeline audits.
Engineered features across categorical (transaction type), text-based (note embeddings), and statistical (amount distribution) dimensions, improving
model input quality by 24%, reducing classification errors, and enhancing downstream machine learning performance.
Trained transformer models (DeBERTa-V3, Longformer) for multi-class transaction classification and document segmentation, optimizing context
windows and tokenization, raising clause and transaction detection accuracy from 77% to 92%.
Fine-tuned LLMs on regulatory, KYC, and AML datasets using LoRA and PEFT, deploying via Docker in secure cloud environments, aligning with SOC 2,
GLBA, and internal IT risk standards.
Implemented RAG pipelines with LangChain, FAISS, and AWS Lambda for real-time retrieval of financial insights, built dashboards with Streamlit and
Plotly, and established feedback loops boosting model relevance by 31% quarterly.

PythonSQLPostgreSQLDeBERTa-V3LongformerDocker+7

ibm research & university of maryland baltimore county

Research Assistant

Feb 2024 – Dec 2024 · 10 mos · Baltimore, MD

Investigated whether pluralistic alignment in LLMs arises from memorization by conducting technical research on the IMHI
dataset (6,000+ mental health instructions) using HPC and ADA servers for large-scale analytics and model evaluation.
Engineered and implemented ETL pipelines to process 6,000+ records, ensuring seamless data ingestion, cleaning, and
transformation for efficient model assessment.
Collected, analyzed, and interpreted data from multiple LLM outputs to evaluate the relationship between memorization and
pluralistic alignment.
Collaborated with cross-functional researchers to integrate insights from structured data enabling data-driven understanding of
memorization effects in pluralistic LLM alignment.
PlanForge, a modular thought-of-search pipeline, was designed to parse tasks described in natural language into structured representations. It combines components using a unit-test feedback loop and carries out a validator-guided breadth-first search over candidate programs and plans. An ablation suite and evaluation tool were developed to identify the roles of tests, feedback, and search.
PlanForge solved 121/130 tasks (92.3%) across 13 classical planning domains (blocksworld, ferry, floortile, visitall, grippers, grid, rovers, logistics, etc), outperforming the strongest best-of-n LLM baseline at 112/130 (86.2%).

HPCADA serversETL pipelinesTableauPlotlyNatural Language Processing (NLP)+1

Mastercard

AI/ML Engineer

Jun 2021 – Jul 2023 · 2 yrs 1 mo · India

Directed the “Dynamic Transaction Insights” project to implement real-time, personalized payment recommendations using Python, SQL, and Spark,
increasing transaction volume by 18% through collaboration with product, analytics, and regional operations teams.
Designed ETL pipelines from MySQL, Hive, and streaming sources using PySpark and Airflow. Engineered features including one-hot encoding,
geohash clustering, and session-level signals. Documented pipelines for scalable, cross-team usage across platforms.
Developed ML models including XGBoost, LSTM, and ARIMA to predict transaction demand, fraud likelihood, and customer churn, enabling insights
across 3M+ users and improving operational efficiency in high-transaction zones by 24%.
Optimized models using Grid Search and Bayesian tuning with time-series and stratified 5-fold validation. Achieved RMSE 7.2, Precision@3 78%, and
AUC 0.88, surpassing legacy systems and strengthening deployment confidence.
Deployed models with Docker and Kubernetes into Mastercard microservices with REST APIs. Integrated Kafka for real-time inference, cutting
latency by 30% while maintaining 99.5% uptime during peak transaction periods in metro regions.
Automated bi-weekly retraining pipelines triggered by data drift using Airflow. Implemented monitoring with Prometheus and Grafana, maintaining
model freshness and 92%+ SLA compliance, while supporting A/B testing of personalized offers and dynamic payment features.