AI/ML engineer

Arshpreet

Python & AI/ML Engineer

Celina, TX · Green Card Holder

Email: ************@*****.***

SUMMARY

Experienced AI/ML Engineer having around 10 years of experience with end-to-end expertise across GenAI, RAG architectures, LLM microservices, NLP pipelines, predictive analytics, and full-stack Python development, delivering solutions for BFSI, Retail, Healthcare, and Manufacturing domains.

Expertise in building secure and compliant AI systems, including FINRA/SEC/SOX-aligned RAG pipelines, PHI/PII redaction, model governance workflows, LLM guardrails, evaluation frameworks, and audit-ready documentation.

Hands-on in developing production-grade ML microservices using FastAPI, LangChain, FAISS, MLflow, PySpark, and AWS/GCP/Azure, with strong proficiency in vector search, embedding refresh pipelines, latency optimization, and LLM cost reduction (ONNX, batching).

Experienced in delivering GenAI capabilities for real business use cases such as retail search, catalog intelligence, customer insights, risk review, case summarization, analyst productivity, and internal support assistants.

Expertise in NLP and document intelligence, including Whisper-based STT, OCR pipelines (Textract, Form Recognizer), clinical entity extraction, ICD-10/SNOMED mapping, rule-based validation, and data quality enhancement for regulated datasets.

Hands-on full-stack Python developer skilled in Django, FastAPI, React, Streamlit, Dash, API security, RBAC, audit logging, workflow automation, and integration with large enterprise systems.

Experienced in ML & data engineering workflows, building Spark/PySpark pipelines, ETL jobs, schema validations, feature stores, data-drift monitoring, and fault-tolerant processing layers.

Expertise in domain-specific analytics including BFSI customer behavior modeling, retail ranking models, clinical data automation, and IoT predictive maintenance using time-series ML (LSTM, Prophet), OpenCV defect detection, and sensor-level feature engineering.

Hands-on with cloud-native AI operations including VPC-isolated deployments, IAM hardening, private endpoints, OpenTelemetry tracing, model lineage tracking, and production monitoring.

Experienced in cross-functional collaboration, working closely with compliance SMEs, clinicians, merchandisers, mechanical engineers, QA teams, and cloud architects to validate outputs, improve accuracy, and deliver business-aligned AI features.

CORE STRENGTHS

AI GenAI & LLMs-- RAG Pipelines, LangChain, Prompt Engineering, Guardrails, FAISS, Embeddings, Whisper STT, LLM Evaluation, Vector Databases (FAISS, Pinecone, Vertex Matching Engine)

NLP & Document Intelligence-- spaCy, Text Classification, OCR (AWS Textract, Azure Form Recognizer), ICD-10/SNOMED Mapping, NER Pipelines

Machine Learning-- Time-Series Forecasting (LSTM, Prophet), Classification Models, Segmentation Models, Anomaly Detection, Feature Engineering

Backend & API Development-- FastAPI, Django, Flask, REST APIs, RBAC, Authentication, Audit Logging, Microservices

Full Stack (UI)-- React (basic), Streamlit, Plotly Dash

Data Engineering & Pipelines-- PySpark, Kafka, Azure Data Factory, ETL Workflows, Data Validation, Feature Store, Data Drift Monitoring

Cloud & MLOps-- AWS, GCP, Azure, Docker, MLflow, Model Registry, VPC Isolation, IAM Policies, ONNX Optimization, OpenTelemetry, Vertex AI · BigQuery · Cloud Functions · VPC-SC · IAM Controls, S3, Lambda, EC2, Glue, EKS/ECS, IAM, KMS, CloudWatch

Security & Compliance-- PHI/PII Masking, HIPAA, FINRA/SEC/SOX Controls, Governance Documentation, Model Risk Management (MRM)

Tools & Dashboards-- Grafana, Kibana, Neo4j, Cypher Queries, SQL, Pandas

EXPERIENCE

Senior AI/ML Engineer — Broadridge, USA June 2024 – Present

Led GenAI and RAG capability development for brokerage compliance, case review, and analyst productivity platforms.

Guided safe AI adoption by working closely with Compliance, Cybersecurity, Data Privacy, and Cloud Governance teams to meet FINRA/SEC/Reg-SCI expectations.

Designed and implemented secure RAG pipelines using LangChain with document-level access control, data-residency rules, and regulated-data filtering.

Developed FastAPI-based ML microservices supporting retrieval, embedding generation, summarization, and structured output generation for compliance teams.

Built evaluation frameworks including grounding checks, hallucination scoring, content-sensitivity rules, and evidence-linking — validated with compliance SMEs.

Developed Whisper-based transcription workflow to convert analyst voice notes into structured text for risk scoring and case documentation.

Implemented MLOps best practices on GCP including model deployment, CI/CD integration, lineage tracking, monitoring, and drift detection.

Designed and implemented agentic workflows combining Gemini models with enterprise tools, APIs, and regulated data sources using Vertex AI and LangChain.

Implemented PHI/PII redaction, encrypted storage, token-level masking, and audit logging aligned with FINRA/SOX/SEC controls.

Integrated MLflow Model Registry for versioned deployments, model lineage, experiment tracking, and change-control documentation required for model approvals.

Integrated FAISS vector search with scheduled embedding refreshes, data validations, and controlled reindexing windows for regulated data pipelines.

Conducted shadow testing and parallel validation of RAG workflows before production rollout.

Worked with the Model Risk Management (MRM) team to prepare model approval artifacts, validation documentation, and lifecycle records.

Deployed FastAPI-based model serving on AWS EKS and ECS with autoscaling, service mesh integration, and fault-tolerant routing.

Implemented S3-backed vector refresh pipelines with lifecycle rules, versioning, and integrity checks for embedding data.

Configured AWS Lambda functions for scheduled embedding generation, metadata cleanup, and real-time compliance rule checks. Used AWS Glue + PySpark to process large compliance datasets feeding downstream RAG and agent-memory pipelines.

Deployed Gemini-powered retrieval and summarization agents using Vertex AI Pipelines with automated approvals, rollback rules, and audit logging.

Implemented LLM guardrail checks including toxicity filtering, compliance-sensitive terminology detection, and safe-response controls.

Contributed to incident triage and L1/L2 analysis for AI system failures, hallucinations, and compliance escalations.

Used Camunda workflow engines to automate compliance checks, review approvals, and human-in-loop escalation paths. Supported audit reviews by preparing evidence logs, lineage data, and governance artifacts for FINRA/SOX reviews.

Added OpenTelemetry traces to monitor LLM latency, request failures, token consumption, and retrieval performance across distributed services.

Optimized inference cost using ONNX quantization, request batching, and GPU utilization tuning, Reduced retrieval latency by optimizing index structure, caching frequent lookups, and trimming low-value embeddings.

Collaborated with cloud architects to enforce VPC isolation, private endpoints, IAM boundaries, and network policies for LLM/ML microservices.

Worked with compliance analysts and product owners to define gold datasets, validate model outputs, tune prompts, and maintain business-acceptable accuracy thresholds.

Authored documentation for model governance, change management, exception handling, and operational readiness required for internal audits and SOC reviews.

Supported incident review workflows by analyzing inference failures, identifying root causes, and recommending mitigation strategies to reduce compliance risk.

AI/ML Engineer — Tredence, USA August 2023 – May 2024

Led development of GenAI and NLP capabilities for retail search, product discovery, and BFSI customer insights platforms.

Designed RAG pipelines with domain-tuned retrieval logic, guardrails, and structured prompting for e-commerce and financial use cases.

Developed LLM-powered product ranking, query understanding, and summarization APIs using FastAPI.

Developed customer-behavior segmentation models used for personalization and targeted search ranking.

Implemented safe-response policies, PII masking, and retrieval access rules based on BFSI compliance standards.

Developed classic ML components (classification/segmentation models) to complement LLM-based retail analytics. Integrated ML workloads with AWS/GCP compute, Docker containers, and Databricks Feature Store for versioned features.

Implemented monitoring dashboards to track drift, retrieval latency, embedding quality, and customer search KPIs. Built FAISS-based vector stores with scheduled refresh workflows to improve embedding relevance and retrieval consistency.

Built Whisper-based speech-to-text workflows to convert support call logs into structured text for downstream NLP models.

Built catalog data-quality workflows including deduplication, attribute consistency checks, and product-title normalization.

Evaluated LLM outputs using BLEU/ROUGE/semantic-similarity metrics for retail and BFSI datasets.

Delivered ML/NLP components across multi-client consulting engagements with staggered releases and SLA-based delivery.

Created PySpark pipelines for large-scale ETL, feature transformations, and input data validation for ML training workflows. Conducted A/B experiments to compare retrieval models, ranking quality, and conversion impact across retail search flows.

Worked closely with retail merchandisers, supply-chain SMEs, and data scientists to align ML features with business rules. Collaborated with BFSI domain teams to validate customer behavior segmentation and reduce hallucinations in customer-facing summaries.

Partnered with QA, cloud engineering, and product teams to prepare UAT plans, deployment checklists, and incremental rollouts.

Documented ML workflows, prompt configurations, and evaluation processes for engineering handoff and client audits.

Full Stack Python Developer — Johnson & Johnson, India March 2020 – December 2022

Developed backend services using Django and FastAPI for clinical data review, workflow automation, and document routing systems.

Designed Camunda-based workflow flows to manage multi-step clinician review cycles and automation triggers. Built secure REST APIs integrating with internal medical systems, document repositories, and clinician review portals.

Implemented authentication, authorization, RBAC, and audit logging to meet healthcare compliance standards.

Developed NLP pipelines for medical entity extraction, diagnoses/medication identification, and ICD-10/SNOMED standard mapping.

Improved NLP accuracy using rule-based validation, domain dictionaries, and custom spaCy components.

Added retry, fallback, and exception-handling logic to stabilize OCR/NLP pipelines in production.

Worked with limited FHIR/HL7 data formats where required for interoperability use cases.

Added performance tuning for backend APIs (query optimization, caching, pagination) to improve clinician workflow speed.

Added basic model-monitoring scripts/logs for tracking data drifts and extraction quality issues.

Highlighted clinical automation outcomes such as reduced reviewer workload and improved case throughput. Built preprocessing logic to clean OCR output, standardize clinical notes, and enhance data quality for ML models.

Integrated Azure Form Recognizer and AWS Textract to automate extraction of lab reports, forms, and physician notes.

Developed validation scripts to compare OCR output with ground truth and flag discrepancies.

Built annotation helpers to support medical reviewers, reduce rework, and accelerate labeling workflows.

Built dashboards using Streamlit and Plotly Dash to visualize extracted medical entities, case counts, and reviewer workloads.

Developed lightweight React components for clinician-facing interfaces as part of full stack duties.

Implemented HIPAA-compliant anonymization and PII/PHI masking pipelines before model training.

Ensured secure data flow using encryption, token-based API access, and endpoint restrictions.

Integrated Azure Data Factory ETL pipelines with downstream Python services for document ingestion and batch processing.

Worked with data engineering teams on schema consistency, validation rules, and data governance for clinical pipelines. Integrated Neo4j to model symptom–treatment–patient relationships for downstream clinical insights.

Wrote Cypher queries to generate relationship-driven views used by clinical analytics teams.

Worked directly with clinical SMEs, regulatory teams, QA analysts, and product managers to validate outputs and refine workflows.

Supported sprint planning, UAT cycles, and cross-team reviews for new automation features.

Documented design decisions, validation logic, and compliance considerations for internal audits.

Python Developer — Cyient, India January 2017 – February 2020

Developed ML models for predictive maintenance using time-series approaches (Prophet, LSTM) on IoT sensor data.

Built Python + PySpark ingestion pipelines for batch and near real-time processing of manufacturing telemetry.

Worked on Kafka-based streaming to process machine vibration, temperature, and operational signals.

Supported root-cause analysis sessions with mechanical teams to interpret anomaly signals and refine model thresholds.

Added edge-device considerations for sensor noise, missing data, and data-drop patterns from factory equipment. Highlighted KPI improvements such as reduced false alarms and improved anomaly detection precision.

Created OpenCV-based visual inspection scripts to detect defects in industrial components. Developed and deployed Flask APIs to surface predictions, alerts, and anomaly summaries to factory dashboards.

Implemented sensor data validation, noise filtering, schema checks, and preprocessing workflows to improve ML reliability. Engineered features for vibration and sensor analytics, including lag windows, rolling statistics, FFT-based features, and trend indicators.

Assisted in ETL workflows combining Spark + Python to support training datasets, model refresh pipelines, and data quality monitoring.

Built Grafana/Kibana dashboards displaying anomaly trends, equipment health, and model outputs for plant supervisors.

Collaborated with mechanical engineers, maintenance teams, and data engineering to validate model signals and adjust features based on machine behavior.

Documented model outputs, assumptions, sensor-level issues, and deployment findings for engineering reviews.

Birlasoft, India August 2014 – December 2016

Developed Python automation utilities to streamline daily reporting, data cleanup, and operational tasks.

Built backend components and REST endpoints using Flask and Django for internal enterprise applications.

Created Pandas + SQL pipelines to generate recurring reports and validate incoming operational data.

Built simple NLP and text classification modules using NLTK and scikit-learn for internal document tagging. Developed ETL and file-processing scripts for scheduled ingestion, validation, and transformation workflows.

Added basic performance tuning for Flask/Django APIs (query optimization, caching strategies).

Added documentation support for automation scripts, data flows, and internal APIs.

Added small contributions to requirement analysis and effort estimation with senior engineers.

Added logging, error handling, and unit tests to improve reliability of internal automation jobs.

Worked with QA and business teams to refine requirements, debug issues, and improve tool usability.

EDUCATION

Master of Computer Science — PTU, India (2014)

Bachelor of Computer Science — PTU, India (2012)

Contact this candidate