Preetham Reddy Vinnamala

About Preetham Reddy Vinnamala

Passionate about leveraging data to drive insightful decisions and create meaningful impact. As a dedicated Master of Science in Data Science candidate at Indiana University with a solid foundation in AI, Statistics, and Advanced Database Concepts, I thrive on tackling complex challenges in the realm of data. My journey encompasses hands-on experiences at Mu Sigma, honing my skills in machine learning, predictive maintenance, and cutting-edge technologies. From developing hyper-automation applications to spearheading innovative video analytics projects, I\'m driven to make a difference through data-driven solutions. Excited to connect, collaborate, and explore opportunities in the world of data science and beyond. Let\'s innovate together!#DataScience #MachineLearning #ProblemSolver\"* Critical Skills: proactive learner, solution and experiment Designer, Problem-solving and analytical skills, Team Player* Programming Languages: R, Python, SQL, MATLAB, Shell Scripting* Databases: Mongo DB, Neo4j, MySQL, NoSQL, Postgres, Elasticsearch, Hive, BigQuery, Alfresco* Tools and IDE: Eclipse, Git, Tableau, Power BI, Shiny R, MS Office, Looker Studio, VS Code* Technologies: Docker, Kubernetes, Linux, RESTful APIs, Spark, Data Structures, TensorFlow, GCP, Natural Language Processing-NLP, Computer Vision, LLM, MLflow, DVC, AWS EC2, GANs, Transformer, RAG.

Experience

1. Ai Engineer

   KPMG US

   Dec 2024 — Present · Houston, TX, US

   Domain-Tuned LLMs & SLM ResearchFine-tuned LLaMA-3.2 on Oil & Gas technical data (SOPs, logs, incident and maintenance reports) to build a high-precision Small Language Model (SLM) for domain QA and document understanding. Used LoRA and Transformer-2.0-style optimizations to improve reasoning efficiency and reduce inference cost, achieving ~35–40% improvement in relevance and hallucination reduction.* Autonomous Audit AI for MESPsBuilt a self-driving audit system for expense vouching, accruals, and SURL using reasoning-based LLMs. Replaced static golden instructions and templates with dynamic procedure generation, human-in-the-loop validation, and missing-evidence detection. Auditors upload documents and sample files; the system determines the procedure and generates MESP-compliant outputs, delivering 60–70% time savings and 40% lower storage costs.* Multi-Agent Financial Market IntelligenceDeveloped a multi-agent AI platform to analyze technical indicators, news, sentiment, and macro signals. A consensus engine produced high-confidence rally and risk signals and scenario simulations, improving signal quality by ~30–40% over single-model systems.* Context Engineering for Long-Horizon AIBuilt a context-engineering platform integrating Redis, SQL, GraphDB, and vector stores with mathematical relevance scoring and context-window extension for open-source LLMs, enabling multi-step reasoning with ~45% faster and ~30% more accurate responses.* RAG & Model OptimizationFine-tuned domain-specific embeddings to replace generic RAG embeddings, improving retrieval by 20–25%(MRR, Recall@10, nDCG). Enhanced SLM reasoning and grounding using QLoRA and GRPO, significantly reducing hallucinations.

Education

• Indiana University Bloomington

  Master's degree, Data Science

• Sri Chaitanya Narayana Junior College

  Intermediate

• National Institute of Technology Raipur

  Electronics and Telecommunication Engineering

  2015 — 2019