Hello, Sreekaran here! đź‘‹I try to investigate the the footprints businesses leave in their data.

From school fee records to construction costs, every dataset I’ve worked with carried a trail: overspending, missed revenue, inconsistent decisions, delayed progress, or hidden risk. I follow those trails with SQL, Python, Tableau, Machine Learning, and Applied AI, while continuously learning new tools that help me work sharper, faster, and closer to the real problem. My goal is to turn scattered records into dashboards, prediction models, and insights that help teams see what their data has been trying to say.

View ResumeContact Me
Sreekaran Reddy

Featured Work

A selected collection of applied AI, data science, cloud, and MLOps projects.

Applied AI

Genre-Controlled Story Generation using QLoRA

Built a reproducible instruction-tuning pipeline for genre-controlled short story generation using Gemma 3-1B, QLoRA, and PEFT, then evaluated whether fine-tuning improved controllability and output quality.

PythonPyTorchHugging FacePEFTQLoRAGemma 3-1B
View Details →GitHub →

Data Science

MedPredicts: Hospital Readmission Forecasting

Built an analytics-driven clinical decision support system to understand hospital readmissions, explain risk drivers, and support proactive staffing and follow-up planning.

PythonPandasXGBoostData CleaningExploratory AnalysisRAG
View Details →GitHub →

MLOps

Vehicle Insurance Eligibility Prediction & MLOps Pipeline

Built an end-to-end MLOps pipeline to predict whether a client should be offered vehicle insurance based on personal details, vehicle attributes, and historical claim data.

PythonPandasNumPyMongoDBDVCFastAPI
View Details →

Cloud Application

RecommenderX: Cloud-Based Movie Rating and Recommendation SaaS

Designed a scalable, production-style movie rating and review SaaS where users can rate movies, write reviews, manage watchlists, and receive AI-powered review support.

PythonDjangoPostgreSQLCloud DeploymentAuthenticationGoogle Cloud Storage
View Details →GitHub →

Data Analytics

Student Success Prediction

Identified factors influencing student success using exploratory analysis, feature engineering, and interpretable predictive modeling.

PythonData AnalysisFeature EngineeringExploratory Data AnalysisPredictive ModelingModel Evaluation
View Details →GitHub →

Articles

Narrative-style technical writing on data, machine learning, LLMs, and applied AI.

How I Slowly Understood What’s Really Happening Inside LLMs like ChatGPT

2026 • 6 min read

How I Slowly Understood What’s Really Happening Inside LLMs like ChatGPT

A personal journey into tokenization, probability, and next-token prediction.

I wrote this as a learning trail through LLMs, starting from the first question that bothered me: how does a model even see text? The article walks through tokenization, Byte Pair Encoding, next-token probabilities, inference, and how a base model behaves like a compressed memory of internet text.

Read on LinkedIn →
Bedtime Story: How Text Became Magic

2026 • 6 min read

Bedtime Story: How Text Became Magic

A story-style introduction to how neural networks read text.

I used a bedtime-story format to explain why neural networks cannot directly read words and why text must become numbers first. The article introduces tokens, vocabularies, encoding, and the strange little bridge between human language and machine-readable input.

Read on LinkedIn →
Chapter 6: When Words Became Pieces

2026 • 7 min read

Chapter 6: When Words Became Pieces

The story of Byte Pair Encoding.

This article explains Byte Pair Encoding through a story where characters, fragments, and frequent pairs slowly become a vocabulary. I wanted to show how BPE helps models handle unfamiliar words by breaking language into reusable pieces instead of memorizing every possible word.

Read on LinkedIn →
The Space Between Words

2026 • 8 min read

The Space Between Words

How embeddings and position help models understand context.

I built this article around a simple misunderstanding between Arjun and Meera: “I need space.” From there, I explain input-target pairs, token embeddings, positional embeddings, and why the same word can mean different things depending on where it stands and what surrounds it.

Read on LinkedIn →
The Night Before Attention Was Born

2026 • 10 min read

The Night Before Attention Was Born

A bedtime story about how machines learned to remember.

I used a story between a father and a little girl to explain why attention became such an important idea in AI. The article starts with old encoder-decoder models, memory, vectors, and the problem that eventually made attention feel less like a feature and more like a rescue mission.

Read on LinkedIn →

Experience

A combined view of my analytics, data science, machine learning, applied AI, MLOps, reporting, and technical support experience.

Internships

Industry experience where I worked across analytics, machine learning, reporting, deployment, and documentation, turning messy real-world data into usable systems and decision-ready insights.

Data Analytics • Machine Learning • MLOps

Analyst Tech Intern

Sree Nirman

May 2023 - Apr 2024

  • At Sree Nirman, the problem was not a lack of data. The problem was that construction cost, labor, material, and project records were scattered, inconsistent, and difficult to trust for planning decisions.
  • I started by working with 50K+ construction cost and operations records, cleaning missing values, schema mismatches, outliers, duplicate entries, and inconsistent cost fields using SQL, Python, Pandas, and NumPy. Before building dashboards or models, I focused on making the data usable, structured, and analysis-ready.
  • Once the data foundation was stronger, I analyzed material costs, labor utilization, timeline delays, budget variance, and project-level performance patterns to understand why costs were drifting and where operational inefficiencies were showing up.
  • I built regression-based construction cost estimation models using historical project, labor, material, location, budget, and progress data. I engineered features, compared model performance using RMSE and R², and tracked 10+ MLflow experiments to evaluate which cost drivers actually improved prediction quality.
  • To make the analysis useful beyond the model, I developed Tableau dashboards and reporting views that translated raw construction records into cost trends, budget variance insights, productivity gaps, and operational risk indicators, improving reporting reliability by 30%.
  • I also helped connect the analytics work to deployment by packaging the trained model as a FastAPI inference service, containerizing it with Docker, deploying it on AWS EC2, and supporting CI/CD workflows with GitHub Actions. The project gave me end-to-end exposure across data cleaning, analysis, modeling, reporting, and production-style model serving.

Financial Analytics • Data Science • Applied AI

Data Analyst Intern

Avanthi High School

Apr 2022 - Jan 2023

  • At Avanthi High School, I inherited a familiar institutional data problem: the school had years of financial, academic, and operational records, but no clean structure to turn those records into reliable decisions.
  • I worked with 12K+ fragmented student financial records and 50K+ institutional expense records across fee collections, scholarships, hostel, dining, academics, activities, and administration. The data had duplicate entries, inconsistent fee formats, missing values, subjective scholarship records, and no standardized schema.
  • I built the school’s analytics foundation from scratch using SQL, Python, Excel, and Tableau, cleaning, normalizing, deduplicating, reconciling, and validating fee, scholarship, and expense records. This improved financial reporting accuracy and consistency by 30%.
  • Once the data became reliable, I analyzed fee collections, department spending, budget variance, category-level costs, collection gaps, scholarship patterns, and monthly financial performance to understand where money was leaking and where spending was drifting beyond plan.
  • I performed variance analysis, anomaly detection, spend-pattern analysis, trend analysis, and distribution checks across 50K+ expense records, identifying over-budget categories, abnormal spending behavior, collection gaps, and 10–15% cost-saving opportunities for leadership review.
  • I built recurring Excel reports, Tableau dashboards, and leadership summaries that gave administrators a clearer view of fee collections, department expenses, budget variance, and category-level financial performance, helping move the school from operating loss to break-even within four months.
  • I also worked on the school’s scholarship decision problem, where manual fee scholarships were inconsistent and sometimes influenced by counter-level bias. I validated historical concession labels with principal-approved records before using them for downstream modeling.
  • I engineered academic and financial features from admission test scores, prior-grade performance, fee category, and a 70/30 academic weighting structure, then built a regression-based scholarship estimation workflow to support more consistent and data-informed scholarship decisions.
  • To strengthen the AI-assisted decision workflow, I fine-tuned a GPT-2 model on historical admission and student-response samples with scoring labels, combining model outputs with structured scholarship logic to create a more stable recommendation signal for leadership review.
  • I documented preprocessing steps, label validation logic, model assumptions, dashboard definitions, and reporting workflows so the school could review, repeat, and explain the analytics process instead of depending on scattered manual judgment.

Assistantships

University technical support experience where I supported students, lab operations, reproducibility, documentation, and structured analytical workflows.

Technical Support • Data Workflows • Student Mentoring

Information Services Lab Assistant

University of Missouri-Kansas City

Aug 2025 - May 2026

  • At UMKC, my work starts when something does not behave the way a student expected: a dataset looks wrong, an output does not make sense, or a workflow is hard to reproduce.
  • I mentored undergraduate students on statistics, EDA, data validation, reproducibility, and structured data handling, helping them move from unclear outputs to cleaner workflows where assumptions, data quality, and analysis steps could be checked.
  • I reviewed datasets, assignments, and analytical outputs to identify inconsistencies, missing assumptions, reproducibility gaps, unclear documentation, and validation issues, helping students make their work easier to explain and repeat.
  • I also supported daily university lab operations by assisting users with hardware and software issues, maintaining lab system documentation, and helping keep student-facing technical support workflows consistent, reliable, and easy to follow.

Skills

A practical toolkit across analytics, machine learning, applied AI, cloud deployment, and data product development.

Languages & Analytics

Core tools I use to query, analyze, and structure data.

PythonSQLRExcel

Data & Visualization

Turning raw datasets into dashboards, KPIs, and insights.

PandasNumPyPower BITableauData CleaningEDAKPI Reporting

Machine Learning

Building, evaluating, and explaining predictive models.

scikit-learnPyTorchXGBoostHugging FaceFeature EngineeringModel Evaluation

Applied AI & LLMs

Fine-tuning, evaluating, and building around language models.

PEFTLoRAQLoRABitsAndBytesGroq APIRAGLLM-as-JudgePrompt Engineering

Cloud & MLOps

Deploying models, tracking experiments, and automating workflows.

MLflowDVCFastAPIDockerAWS EC2AWS S3AWS ECRGitHub ActionsGoogle Cloud Storage

Web & Apps

Building usable interfaces and backend systems for data products.

DjangoPostgreSQLMongoDBStreamlitClerk Authentication

Connect with me

© 2026 Sreekaran Reddy. Built to showcase data, machine learning, and applied AI work.