AI Engineer - Kaiser Permanente - Charlotte, NC
HarshMehta
I turn raw data into agentic systems and real decisions. The interesting problems are the ones where the model doesn't get to be wrong quietly.
Experience
Where the systems had to work in front of people.
Kaiser PermanenteAI Engineer (Contract)
Citation-backed RAG over 5,000+ pages of compliance documentation. A 3-month policy review now takes one week.
Open →
ATCAI Engineer
Multi-agent LLM workflows for 7 enterprise use cases: 88% accuracy, hallucinations under 4%, 72% autonomous completion.
Open →
ExternAI Engineer (Externship)
Wayfair and Outamation: 95% extraction on 200-page mortgage files; A100 embedding cut from 120s to 15s.
Open →Projects
Proof of work. Clone them and they run.
Production RAGTriple-OCR-RAG
Three parallel RAG pipelines for mortgage extraction with FAISS and a 5-tier local fallback. 95% accuracy.
Open →Healthcare RAGClaimsRAG-HM
Compliance RAG over 5,000+ pages: hybrid retrieval, GraphRAG, FastAPI, Docker. Sibling of the Kaiser system.
Open →Video RAGVidRag
Ask a whole YouTube course a question, get the answer and the exact second it came from. Browser-only, bring your own key.
Open →About
Translation work, mostly.
Hi, I'm Harsh Mehta. I build workflows that turn raw data into agentic AI systems, and I'm currently going deeper on CUDA and GPU performance. I like learning about systems that scale. I think often about designing systems that help you keep up with your agents.
I care about what makes these systems work in front of people: evaluation loops, guardrails, and the boring observability code that tells you when your agent is drifting.
I also mix front-of-house audio for rooms from 300 people to stadiums, which turns out to be exactly the right training for shipping things that must work live.
Education
University of Wisconsin-MadisonM.S. Information (Data, ML, Cloud)
Started in applied machine learning, then LLMs, data engineering, and cloud. Finished with a 3.95 GPA.
Open →
University of MumbaiBachelor of Management Studies
Undergraduate foundation in business, statistics, and analytics before moving into data science and ML.
Open →Certifications
DatabricksData Engineer Associate
Building and maintaining data pipelines, ETL, and the Lakehouse Platform on Databricks.
Open →
SnowflakeSnowPro Core
Core Snowflake concepts: architecture, data loading, virtual warehouses, security, and performance.
Open →
Tableau / SalesforceTableau Data Analyst
Connecting to, preparing, and analyzing data and building Tableau dashboards.
Open →
MicrosoftPower BI Data Analyst (PL-300)
Preparing, modeling, visualizing, and analyzing data and deploying Power BI deliverables.
Open →
MicrosoftAzure Data Scientist (DP-100)
Designing and running Azure ML workloads: training, deployment, pipelines, and MLOps.
Open →
MicrosoftAzure Fundamentals (AZ-900)
Foundational cloud concepts and core Azure services, privacy, compliance, and pricing.
Open →Skills
The stack I reach for.
AI & LLM Engineering
- LLMs
- RAG (hybrid search, reranking, citations)
- Agentic Systems (LangChain, LangGraph)
- MCP
- Prompt Engineering
- LLM Evaluation
- Guardrails & Responsible AI
- Vector DBs (Pinecone, FAISS, pgvector)
- Embeddings (BGE, Sentence-Transformers)
ML & MLOps
- PyTorch
- scikit-learn
- Transformers
- NLP (spaCy, NLTK, BERT, RoBERTa)
- Predictive Modeling
- Model Monitoring (drift, LangSmith)
- CI/CD for ML (GitHub Actions)
- Docker
Data Engineering
- Spark / PySpark
- Kafka
- Airflow
- dbt
- Snowflake
- Databricks
- ETL / ELT
- BigQuery
- PostgreSQL
- MongoDB
Cloud
- AWS (SageMaker, Bedrock, Lambda, S3, Glue, ETL, Secrets Manager, EC2, EMR, QuickSight, Redshift)
- Azure OpenAI
- GCP Vertex AI
Languages
- Python
- SQL
- R
- TypeScript
- Go
- C++
Analytics & BI
- Tableau
- Power BI (DAX)
- Statistical Analysis
- Streamlit
Web / API
- FastAPI
- React
- Node.js
- REST APIs
Writing
Notes from the unglamorous middle.
EssayBuilding Intelligent Automation Systems
The discipline of small agent counts, and what nobody tells you about the long tail of production failures.
Open →
EssayData Science in Practice
Three projects where the modeling was the easy part, and why data science is mostly translation work.
Open →
EssayThe Future of AI in Business Operations
Four patterns showing up in real AI deployments, and why the value is going to live in the unglamorous middle.
Open →Get in touch
If you are hiring an AI engineer who cares more about month three than the demo, or building something strange enough to be worth shipping carefully, my inbox is open.
Charlotte, NC · Remote-friendly · 608-298-8733