Bhavana Gubala

Generative AI and Machine Learning Engineer with 4 years of experience designing scalable NLP and RAG-based solutions. Skilled in LLM fine-tuning, MLOps, and real-time data systems, with a strong focus on building reliable, compliant, and production-ready AI applications in enterprise environments.

Gen AI Engineer - State Street Corporation - Massachusetts, USA

(2024-09)

• Designed and implemented an end-to-end RAG pipeline using LangChain and Pinecone to process 500K+ financial documents (10-K filings, earnings transcripts, macro reports), enabling sub-3 second semantic search with 94% analyst-validated relevance.
• Fine-tuned a GPT-4-based model using LoRA/PEFT on 18 months of proprietary research data, improving accuracy of domain-specific financial terminology by 31% in portfolio commentary generation.
• Developed a multi-agent orchestration layer with LlamaIndex to break down complex investment queries into structured sub-tasks, reducing analyst turnaround time by 40% for multi-step research workflows.
• Built an output validation framework using Guardrails AI to enforce SEC compliance requirements, reducing hallucinated or noncompliant financial content by 87% and supporting production approval from legal teams.
• Developed and deployed a FastAPI-based inference service, containerized with Docker and managed on Kubernetes (AKS), supporting 200+ concurrent users with 99.7% system uptime.
• Set up an MLflow-based model registry and CI/CD pipeline using GitHub Actions, reducing model deployment time from two weeks to three days while maintaining auditability and version control standards required in financial environments.

Associate Engineer – AI/ML - IBM India - India

(2021-06 - 2023-08)

• Built a multi-label NLP classification model using fine-tuned BERT on 1.2M+ ServiceNow tickets to automatically categorize and route incidents, achieving 91.4% accuracy and reducing MTTA from 47 minutes to 8 minutes.
• Developed an anomaly detection pipeline using Isolation Forest and LSTM autoencoders on infrastructure telemetry (CPU, memory, network), enabling prediction of incidents ~15 minutes before SLA breaches with 88% precision.
• Designed a T5-based sequence-to-sequence model to generate resolution notes and draft RCA reports from historical incidents, reducing Level-2 engineering documentation effort by 60%.
• Implemented a real-time data pipeline using Apache Kafka and Spark Structured Streaming to process 50K+ telemetry events per minute, delivering features to ML models with sub-200 ms latency.
• Deployed scalable ML services on IBM Kubernetes Service (IKS) using Helm, and integrated with Watson OpenScale for model monitoring, bias detection, and automated retraining triggers.
• Partnered with SRE and NOC teams to integrate the platform with PagerDuty and Slack, enabling automated alerting and reducing L3 escalations by 35% over six months.

Master of Science - Computer Science - University of Massachusetts Dartmouth