Open to opportunities

Shivani Ajay Harane

@shivaniajayharane

Message

AI/ML Engineer building production-grade GenAI and RAG systems with LLMOps.

United States

Message

What I'm looking for

I’m looking for a role where I can ship production GenAI/RAG systems end-to-end—hybrid retrieval, LLM fine-tuning, and LLMOps observability—while scaling reliable distributed infrastructure. I want measurable impact, clean tooling, and strong engineering ownership.

I’m an AI/ML Engineer with 5+ years of software engineering experience and 2+ years designing production-grade Generative AI systems. I focus on building end-to-end RAG pipelines that perform reliably—covering everything from embedding strategy and hybrid retrieval to deployment.

In my recent work, I built a production RAG system (FastAPI + LangChain + Cohere Command-R) with sparse BM25, dense ChromaDB retrieval, and cross-encoder reranking. I achieved RAGAS answer relevance of 0.87+ across 100K+ document corpora while delivering horizontally scalable, containerized performance.

I’m equally strong on the “keep it working” side: I instrument full LLMOps observability with Prometheus + Grafana, tracking P95 latency, error rate, and retrieval throughput at 15-second granularity. I’ve also created live A/B testing and rollback for retrieval weights and model selection, plus adaptive document normalization to resolve PDF ingestion failures across heterogeneous corpora.

Before specializing in GenAI, I delivered high-impact backend and distributed systems work—cutting duplicate transactions by 80%, reducing API P95 by 44% using Redis caching, and improving infrastructure uptime by 35% with C#/.NET services and fault-tolerant ingestion patterns. I bring that systems mindset to every AI pipeline I build.

Experience

Work history, roles, and key accomplishments

Current

AI Research Assistant

Current

Binghamton University

Mar 2026 - Present (3 months)

Built a production RAG system (FastAPI + LangChain + Cohere Command-R) with hybrid retrieval (BM25 + ChromaDB) and cross-encoder reranking, achieving RAGAS answer relevance of 0.87+ across 100K+ documents. Deployed end-to-end LLMOps observability (Prometheus + Grafana), implemented zero-downtime A/B testing for retrieval weights/LLM selection, and added adaptive normalization to fix PDF ingestion

fastAPI Lang Chain Cohere ChromaDB Grafana Cross Encoder Re Ranking A B Testing Docker RAG Evaluation (RAGAS)

Machine Learning Intern

Turtle & Hughes

Jun 2024 - Dec 2024 (6 months)

Deployed a stacked ensemble forecasting model (Random Forest + XGBoost + Linear Regression) on Azure, reducing MAPE by 40% versus a single-model baseline and cutting hyperparameter iteration time from 2+ days to under 4 hours. Automated an ETL pipeline (Knime + Java) to eliminate manual workflows across 5+ teams and improved ingestion latency to sub-hour, supporting inventory risk decisions via Ql

Azure ETL KNIME Java Random Forest XGBoost Cross Validation

Software Engineer

Tata Consultancy Services

Nov 2020 - Oct 2022 (1 year 11 months)

Developed an idempotent booking microservice (Java 8 + Spring Boot + PostgreSQL) using availability locking and request deduplication, eliminating duplicate transactions by 80% and reducing manual call volume by 40% under sustained peak load. Reduced API P95 latency from 900ms to 500ms (44%) using Redis caching with adaptive TTL/rate limiting and Kafka-based async event propagation for fault-toler

Java 8 Spring Boot PostgreSQL Redis Apache Kafka Microservices CI CD

Software Engineer

Tata Consultancy Services

Oct 2018 - Oct 2020 (2 years)

Built high-availability C#/.NET services for dispersed data ingestion, improving infrastructure uptime by 35% across 5+ sites, and designed REST APIs/ETL workflows to sync ERP-integrated factory management software for real-time analytics. Owned incident response and root-cause analysis across 5+ production nodes, improving resolution practices to maintain SLA compliance for mission-critical servi