Looks like you have JavaScript disabled. For the full Twine experience, you will need to re-enable it.

I'm Josthana Runkana, an AI/ML engineer with a passion for building end-to-end ML systems that scale in regulated environments. I design, train, and deploy production-grade models and LLM-powered pipelines, with a strong focus on governance, reliability, and observability. I enjoy turning complex data challenges into robust, scalable solutions that deliver measurable business impact. Currently, I lead initiatives that combine traditional ML with retrieval-augmented generation, agentic AI, and MLOps to reduce analyst workload, improve decision support, and tighten risk and compliance controls across financial services and enterprise SaaS domains. I’m continuously expanding my expertise in model evaluation, bias mitigation, and latency optimization to ensure responsible, high-performance AI in production.…I'm Josthana Runkana, an AI/ML engineer with a passion for building end-to-end ML systems that scale in regulated environments. I design, train, and deploy production-grade models and LLM-powered pipelines, with a strong focus on governance, reliability, and observability. I enjoy turning complex data challenges into robust, scalable solutions that deliver measurable business impact. Currently, I lead initiatives that combine traditional ML with retrieval-augmented generation, agentic AI, and MLOps to reduce analyst workload, improve decision support, and tighten risk and compliance controls across financial services and enterprise SaaS domains. I’m continuously expanding my expertise in model evaluation, bias mitigation, and latency optimization to ensure responsible, high-performance AI in production.

Josthana Runkana

AI Engineer, Data Scientist, Full Stack Developer, +2





I'm Josthana Runkana, an AI/ML engineer with a passion for building end-to-end ML systems that scale in regulated environments. I design, train, and deploy production-grade models and LLM-powered pipelines, with a strong focus on governance, reliability, and observability. I enjoy turning complex data challenges into robust, scalable solutions that deliver measurable business impact. Currently, I lead initiatives that combine traditional ML with retrieval-augmented generation, agentic AI, and MLOps to reduce analyst workload, improve decision support, and tighten risk and compliance controls across financial services and enterprise SaaS domains. I’m continuously expanding my expertise in model evaluation, bias mitigation, and latency optimization to ensure responsible, high-performance AI in production.…I'm Josthana Runkana, an AI/ML engineer with a passion for building end-to-end ML systems that scale in regulated environments. I design, train, and deploy production-grade models and LLM-powered pipelines, with a strong focus on governance, reliability, and observability. I enjoy turning complex data challenges into robust, scalable solutions that deliver measurable business impact. Currently, I lead initiatives that combine traditional ML with retrieval-augmented generation, agentic AI, and MLOps to reduce analyst workload, improve decision support, and tighten risk and compliance controls across financial services and enterprise SaaS domains. I’m continuously expanding my expertise in model evaluation, bias mitigation, and latency optimization to ensure responsible, high-performance AI in production.

Available to hire

I’m Josthana Runkana, an AI/ML engineer with a passion for building end-to-end ML systems that scale in regulated environments. I design, train, and deploy production-grade models and LLM-powered pipelines, with a strong focus on governance, reliability, and observability. I enjoy turning complex data challenges into robust, scalable solutions that deliver measurable business impact.

Currently, I lead initiatives that combine traditional ML with retrieval-augmented generation, agentic AI, and MLOps to reduce analyst workload, improve decision support, and tighten risk and compliance controls across financial services and enterprise SaaS domains. I’m continuously expanding my expertise in model evaluation, bias mitigation, and latency optimization to ensure responsible, high-performance AI in production.

Skills

Experience Level

Expert

Expert

Expert

Expert

Expert

Expert

Expert

Language

English

Fluent

Work Experience

AI/ML Engineer at Morgan Stanley

December 1, 2024 - Present

Architected and deployed a multi-class customer risk analytics platform using XGBoost and LightGBM ensembles with SHAP-based explainability, plus an anomaly-detection layer auditing 50M+ daily transactions. Applied SMOTE and imbalanced-data techniques to achieve 91% precision while reducing manual risk reviews by 35%. Developed a LambdaMART ranking model for personalized wealth advisory recommendations by engineering 80+ features from client transactions, portfolios, and market indicators, improving NDCG@10 by 22% and increasing product adoption by 14%. Implemented an Agentic AI platform for financial document intelligence with LangGraph, tool calling, semantic search, and RAG; established model governance, model risk management, bias evaluation, and SR 11-7 compliance controls, reducing analyst review effort by 65%. Fine-tuned a domain-adapted LLM on 2M+ financial documents using LoRA/QLoRA with DeepSpeed distributed training across 8× GPUs; integrated SpaCy-based NER workflows to im

ML Engineer at Freshworks

August 1, 2022 - July 1, 2023

Shipped a BERT-based multilingual intent classification and sentiment analysis platform across 60+ support categories, achieving 93% accuracy, reducing ticket resolution time by 38%, and improving CSAT by 18 points across 65,000+ customers. Built a multilingual NER pipeline using RoBERTa and SpaCy, deployed via FastAPI services with automated monitoring, processing 2M+ customer interactions monthly at sub-120ms latency. Led PoC across 12+ routing models, applying feature engineering, SMOTE, and imbalanced-data handling techniques to improve robustness while reducing feature volume by 60%. Developed an evaluation framework using DeepEval and TruLens to monitor model quality, hallucination risk, and business KPIs, identifying improvements that increased CSAT by 12%. Accelerated data preparation 10× using Spark, Kafka, and Dataproc pipelines, reducing processing time for a 50M-record multilingual corpus from 14 hours to 90 minutes while supporting shared feature-store workflows. Mentored

ML Engineer at KPIT Technologies

January 1, 2021 - July 1, 2022

Delivered a production computer vision defect detection system (ResNet-50 + targeted augmentation) for an automotive OEM, raising detection precision from 76% to 94%, reducing false-positive escalations by 60%, and deploying via REST API across 3 manufacturing plants. Engineered LSTM-based time-series forecasting models for predictive maintenance across 3 manufacturing sites achieving 80% failure-recall at a 7-day horizon, enabling proactive intervention that reduced unplanned downtime by up to 22%. Reduced edge model footprint 55% and inference latency from 180ms to 55ms via INT8 quantization, structured pruning, and ONNX Runtime export to meet a 100ms production SLA on NVIDIA Jetson devices across 3 deployments with zero holdout accuracy regression. Automated end-to-end feature engineering and data validation pipelines (Python, Pandas, DVC) for 5M-row/day sensor telemetry stored in Delta Lake, eliminating 20 hours/week of manual preprocessing and enabling full data lineage tracking f