Neha Dabeeru · Software Engineer

01 · About

Software Engineer with 4+ years building production systems across both the backend/API layer and the data/ML layer: GraphQL services and event-driven workflows on one side, PySpark pipelines, GenAI retrieval, and MLOps on AWS on the other.

At Merck, I work across a clinical data platform used by 2K+ users: AppSync GraphQL APIs and async event-driven workflows, alongside PySpark/Databricks ETL pipelines processing 500GB+ per cycle with CDC and governed Delta Lake layers. GenAI-assisted encoding (hybrid BM25 + neural k-NN, Bedrock embeddings) hits 99.5%+ accuracy and cut manual review by 55%. Earlier, at Bank of America, I built backend and migration tooling for financial-risk and Cards & Payments systems: reconciliation engines, event-triggered pipelines, and distributed ETL across 20+ risk pipelines.

I'm interested in systems where data engineering, backend reliability, and applied AI meet: pipelines that feed production APIs, retrieval systems backing GenAI features, and the reliability work (idempotency, state management, failure recovery) that keeps both running.

02 · Experience

Where I've worked

Jan 2026 – Present
MSD

Software Engineer / Application Lead
Merck Sharp & Dohme · Pennsylvania, USA

Clinical Encoding Platform
- CDL (Encoding Engine) is Merck's clinical data management platform, used to standardize verbatim terms into MedDRA/WHODrug codes across clinical trials for 2K+ users globally on a 3M+ req/day platform.
- Built PySpark/Airflow ETL pipelines with CDC, schema evolution, and merge/upsert logic over Databricks and Delta Lake, processing 500GB+ per cycle across incremental and nightly full loads.
- Built AppSync GraphQL APIs with Lambda and VTL resolvers for encoding, approval, audit, and external-partner workflows.
- Redesigned long-running AppSync operations as async workflows using DynamoDB Streams, Step Functions, and GraphQL subscriptions, eliminating AppSync timeouts and cutting Databricks latency from 60s to 2s.
- Built a hybrid BM25 + neural k-NN scoring pipeline with Bedrock embeddings across 20+ OpenSearch indices, reaching 99.52% top-match accuracy and cutting manual review 55%.
- Established governed Delta Lake layers with Unity Catalog and Immuta, enforcing row/column-level masking and role-based access for sensitive clinical data.
AppSync DynamoDB job Streams Step Functions Databricks/Lambda GraphQL subscription UI

Specification Management System
- Led the Specification Management System end-to-end: replaced a manual, Excel-based clinical-trial encoding workflow with a governed Draft → Publish → Promote application with versioning, audit trails, and role-based access, reducing manual handoffs and approval turnaround by 60%.
Reliability & Production Engineering
- Implemented append-only DynamoDB audit logging with Global Tables replication, tracking user actions and state changes for 2K+ users.
- Raised backend test coverage to 90%+ with a pytest and moto-based AWS test suite, preventing regression of a production data-integrity defect.
- Reduced repeat production incidents through on-call triage, root-cause analysis, postmortems, and permanent fixes across APIs, workflows, and data pipelines.
Impact
- 2K+ users across the platform
- 99.52% top-match auto-coding accuracy (98.46% on secondary matches)
- 5–6 hour incremental sync for updated clinical data
AppSync/GraphQLDynamoDBStep FunctionsDatabricksMuleSoftpytest
Jan 2025 – Dec 2025
CSO

Application Developer
CS Optima · Texas, USA
- Cut clinical-study onboarding from 8–10 weeks to ~1 week by re-architecting manual onboarding into async validation, standardization, and ingestion pipelines.
- Architected a RAG-based schema-mapping pipeline with pgvector semantic retrieval, AWS Bedrock reranking, and human review, using deterministic structure fingerprints to reuse approved mappings for recurring schemas.
- Implemented data-quality and reconciliation controls for schema mismatches, duplicates, discrepancies, and idempotent re-ingestion of corrected data.
- Modeled Neo4j-backed data lineage across source records, transformations, and corrections, enabling row/cell-level provenance and impact analysis.
RAGpgvectorBedrockNeo4jData Lineage
Jun 2024 – Aug 2024
BoA

Software Engineer II (Contract)
Bank of America · North Carolina, USA
- Modernized legacy codebase using AI-assisted refactoring, consolidating duplicate logic into reusable modules and expanding automated test coverage.
- Built checksum-based reconciliation engine to detect schema drift and silent data mismatches across Cards & Payments DB migration.
- Automated cross-database validation (referential integrity, data-type, record-level checks), cutting manual pre-cutover testing.
- Generated structured exception reports and failure summaries, accelerating migration triage, remediation, and prod cutover decisions.
Data ReconciliationSQLMigration Testing
Jun 2020 – Jan 2023
BoA

Software Engineer I
Bank of America · Tamil Nadu, India
- Delivered Python backend services integrating trading systems and SQL databases for market and credit-risk data calculations and regulatory reporting.
- Improved regulatory reporting accuracy to 99.5% with automated reconciliation detecting missing records, balance mismatches, and data-quality issues.
- Reduced risk-batch latency 40% by parallelizing Python workloads and tuning complex queries, indexes, and database access across high-volume datasets.
- Reduced infrastructure costs 35% by redesigning always-on workflows into event-triggered pipelines, eliminating idle resource time.
- Built backend microservices and REST APIs for data ingestion, validation, and processing, plus reusable internal API components (connection pooling, transaction management, retry logic) adopted across downstream teams.
- Created a rule-based, multi-dialect SQL translation microservice, reducing manual migration effort by 80%.
- Strengthened pipeline reliability by making processing steps idempotent with checkpoint/restart recovery, preventing duplicate processing on failure.
SparkEvent-Driven PipelinesREST APIsBitbucket CI/CDSQLIdempotency
Jul 2017 – Sep 2017
RINL

Software Engineer Intern
Rashtriya Ispat Nigam Limited · Andhra Pradesh, India
- Developed Python services for ERP-integrated workflows, automating validation and processing of 10K+ manufacturing and operational records/day.
- Created a database-driven web application with reusable business logic and server-side validations, supporting 500+ users across enterprise workflows.
PythonERPWeb Application

03 · Projects

Selected engineering projects

MedPredict

shipped

Hospital readmission prediction system with LLMs

Dataset & model pipeline

100GB of MIMIC-III clinical data (40K+ records), processed through PubMedBERT, ClinicalXLNet, and BigBird transformer models for embeddings and classification.

Clinical record Transformer embedding FAISS retrieval PyTorch classifier FastAPI inference API AWS deployment

Retrieval & serving

FAISS retrieval feeds a PyTorch classifier for readmission risk; a separate GPT-4-backed RAG prototype surfaces actionable clinical insights on top of the prediction, improving decision-making efficiency by ~80%.

Deployment

Full-stack, Dockerized (Angular frontend, FastAPI backend, Firebase auth, AWS S3 for clinical data storage), deployed with HIPAA-compliant handling end-to-end.

Evaluation & what I built

92%+ readmission-risk accuracy at ~2s inference latency. I personally implemented the FAISS retrieval layer, the FastAPI serving path, the Docker/AWS deployment, and the RAG prototype integration.

PyTorchFAISSGPT-4 / RAGAngularFastAPIDockerAWS S3HIPAA

Devpost ▶ Demo video GitHub

SFJobSync+

Sep–Dec 2023

Real-Time Workforce Data Platform · San José State University

Pipeline

Built a big-data pipeline and web app for SF County employees on AWS, streaming real-time workforce data with dual Kafka producers and PySpark transformations into Redshift/S3 storage.

App & serving

Built REST APIs and a React application on top of the pipeline for downstream analytics and prediction workflows.

Evaluation

Logistic Regression model reached 94% accuracy; Redshift and Tableau powered decision-support analytics for SF County stakeholders.

KafkaPySparkAWSRedshiftReactREST APIs

GitHub

04 · Skills

Technical skills

Core strengths

PythonPySparkGraphQLAWS LambdaDatabricksDynamoDBPostgreSQLBedrockRAGAirflow

Additional tools & technologies

Languages

Python, Java, SQL, JavaScript, TypeScript

GenAI & ML

AWS Bedrock (Claude/Haiku), Prompt Engineering, RAG, Vector Search, pgvector, FAISS, PyTorch, Transformer Embeddings, Model Serving

Backend & APIs

REST APIs, GraphQL, FastAPI, AWS AppSync, API Gateway, AWS Lambda, VTL Resolvers, Async Processing, Auth/AuthZ

Distributed Systems

Event-Driven Architecture, Serverless, Workflow Orchestration, Message Queues, Idempotency, Retry Strategies, Failure Recovery, State Management, Optimistic Locking, Concurrency Control

AWS & Cloud

Lambda, Step Functions, EventBridge, SQS, S3, DynamoDB, AppSync, API Gateway, OpenSearch, EMR, EC2, IAM, KMS, CloudWatch, X-Ray

Databases & Search

PostgreSQL, MySQL, DynamoDB, Snowflake, Delta Lake, Amazon OpenSearch, HDFS, Hive, Impala

Data Processing

Apache Spark, PySpark, Spark SQL, Databricks, ETL/ELT, Schema Enforcement, Data Reconciliation, Deduplication

Infrastructure & DevOps

Terraform, Docker, GitHub Actions, CI/CD, Git, IaC, Structured Logging, Monitoring, Distributed Tracing, Production Support

05 · Education & Certification

Education

SJSU

M.S., Data Analytics

San José State University, California

Jan 2023 – Dec 2024

SRM

B.E., Computer Science and Engineering

SRM University, India

Jun 2016 – May 2020

AWS Certified Data Engineer – Associate

Amazon Web Services

Issued May 2025 · Expires May 2028

06 · Contact

Let's talk

Open to Software Engineer, Data Engineer, and Data & AI Platform roles. Based in San Francisco, CA. Open to remote and relocation.

nehadabeeru@outlook.com ndabeeru@outlook.com (alt) +1 (240) 386-0822 LinkedIn GitHub