Skip to content

Open to 2027 Data Science, ML, and AI opportunities

Tempe, Arizona

BDA + AI @ ASU ’27

I turn messy datainto useful decisionsand dependable systems.

Pranshu Ghori in a blue polo and sunglasses, standing in front of the Golden Gate Bridge on a clear day

Pranshu GhoriData Science Co-op, Command Credit Corp

Data Science Co-op at Command Credit Corp and a concurrent Business Data Analytics and Artificial Intelligence student at Arizona State University. I work across the whole path from raw records to something people rely on: building labels where none exist, testing models honestly, and shipping the parts that hold up.

  • Applied machine learning
  • Evaluation
  • Record linkage
  • Data systems
  • AI applications
  • Statistics

01 — Education

Two degrees, one question: what makes data useful?

Aug 2024 — Aug 2027 (expected)Tempe, Arizona

Arizona State University

  • B.S. Business Data Analytics
  • B.S. Artificial Intelligence

Concurrent degrees, both STEM-designated.

Cumulative GPA
4.00
Expected graduation
Aug 2027
  • Dean's List — Fall 2024
  • Dean's List — Spring 2025
  • Dean's List — Fall 2025
  • Dean's List — Spring 2026

02 — Experience

Where the work has happened.

  1. May 2026 — Present

    Co-op · scheduled through Dec 2026

    Active

    Data Science Co-op

    Command Credit Corp · Scottsdale, Arizona

    Data science across a business-credit product: outcome data, risk modeling, entity resolution, data quality, and analytics.

    • Built an outcome-labelled dataset from public filings and a regularized ranking score that ships in a business-credit product.
    • Designed and shipped cross-provider entity matching with conflict safeguards, readable decision trails, and user confirmation for ambiguous cases.
    • Audited completeness and comparability of commercial credit data and built percentile-based score displays with deterministic explanations.
    • Prototyped a public-records business-data pipeline with record linkage, provenance, and transparent scoring experiments.
    • Analyzed legacy product usage with SQL to inform the design of a replacement product.
    • Python
    • SQL
    • DuckDB
    • Polars
    • scikit-learn
    • TypeScript
    Read the case studies

  2. May 2025 — Aug 2025

    Internship

    Machine Learning Engineer Intern

    Bigscal Technologies Pvt. Ltd. · India

    Supported a machine learning team on modeling workflows.

    • Contributed to data preprocessing, feature engineering, and model training workflows in Python and scikit-learn.
    • Built and tested classification and regression models and helped compare them during model selection.
    • Python
    • scikit-learn

03 — Flagship

The study I would want read first.

01 · Commercial credit-risk modeling

Data Science · Machine Learning

Status: Shipped score · separate research

Build the outcome labels first. Then earn a score.

Tested whether commercial credit signals separate businesses with observed adverse outcomes, then built a regularized ranking score that now ships in a business-credit product.

I built the labelled outcome dataset, screened the available signals, developed and evaluated the model, and packaged the score with reference examples so the product implementation could be tested against it.

  • Python
  • DuckDB
  • NumPy
  • pandas
  • SciPy
  • scikit-learn
  • Jupyter
  • TypeScript
Read case study

How it was checked

  • Untouched held-out test set
  • Temporal, backward, and prospective forward checks
  • Leakage and selection-bias review
  • Python ↔ TypeScript parity on reference examples

What shipped

  • Ranking scoreStatus: Shipped
  • Additional model familiesStatus: Research
  • Overall verdictStatus: Shipped · rule-based

From public filings to a tested score

  1. Public filings

    Filter to business entities

  2. Normalize + link

    Staged name matching; confident links only

  3. Enrich

    Attach business records

  4. Screen signals

    Direction, discrimination, leakage

  5. Regularized model

    Percentile-binned inputs

  6. Reference examples

    Python ↔ TypeScript parity

Simplified flow. Each stage was evaluated before the next consumed its output.

04 — Professional work

Four more studies from the same desk.

Company work is presented through anonymized summaries and illustrative examples.

  • 02 · Data Systems

    Status: Shipped matcher · research prototype

    Cross-provider entity resolution

    Matched the same business across inconsistent names and addresses from multiple commercial data providers, with conflict safeguards, a review path for ambiguous cases, and readable decision trails.

    Technical detail
    • Shared normalizer with parsed street components
    • Edit-distance and n-gram similarity with hard conflict rules
    • Link / review / no-link outcomes with decision trails
    • Server-validated user confirmation for ambiguous cases
    • TypeScript
    • React
    • SQL
    • Python
    • String similarity
    • Record linkage
    Read case study

  • 03 · Data Systems / Machine Learning

    Status: Working local prototype

    Public-record business-data pipeline

    A config-driven local pipeline that combines public records into one business reference, with layered processing, provenance, record linkage, and transparent modeling experiments.

    Technical detail
    • Raw → staging → normalized → identity → export layers
    • Deterministic links first, probabilistic linkage for the rest
    • Linker tuned on known-correct identifier pairs
    • Naive Bayes industry classifier with abstention and self-training
    • Python
    • Polars
    • DuckDB
    • Parquet
    • pytest
    • rapidfuzz
    • YAML
    • Census Geocoder
    Read case study

  • 04 · Data Science

    Status: Analysis delivered · displays shipped

    Credit-data quality and benchmarking

    Audited completeness, comparability, and meaning in commercial credit data, and turned the findings into recommendations and more interpretable score displays.

    Technical detail
    • Seeded reservoir sampling from very large files
    • Sentinel-aware fill rates and distribution analysis
    • Rank correlation, weighted agreement, equipercentile equating
    • Midrank percentile displays with deterministic explanations
    • Python
    • pandas
    • NumPy
    • SciPy
    • scikit-learn
    • Matplotlib
    • DuckDB
    • Jupyter
    Read case study

  • 05 · Data Science

    Status: Analysis and recommendations delivered

    Customer and product analytics

    Analyzed usage and purchasing patterns in a legacy business-data product with SQL, to inform the design of its replacement.

    Technical detail
    • Explicit grain with distinct counts
    • Conditional aggregation, time bucketing, cohort comparison
    • Cross-system reconciliation
    • Recurring analyses automated in Python
    • SQL
    • MySQL
    • DBeaver
    • Python
    • pandas
    • Excel
    Read case study
All case studies

05 — Personal projects

Built on my own time.

Independent projects, separate from any employer work above.

  • AI application

    InsightDesk

    Repository not public

    An LLM analytics assistant that answers executive questions over a SQL Server database, where every number is computed by deterministic SQL and the model only plans approved tool calls and explains the results.

    How it works
    • The model never produces numbers itself: it plans calls to approved tools, deterministic SQL computes every figure, and the model explains what came back.
    • Role-based tool permissions, strict JSON schemas for every tool call, and an audit trail of each request.
    • A claim validator rejects any narrative that cites a number not present in the retrieved evidence.
    • Seven reconciled metric contracts define how each business metric is calculated.
    • Evaluation harness with held-out suites and release gates: scored 62% on blind cases, traced every miss to a defect, reached 16/16 on seen cases after fixes, and benchmarked against a text-to-SQL baseline.
    • Python
    • SQL Server
    • SQLAlchemy
    • Pydantic
    • FastAPI
    • OpenAI / Anthropic APIs
    • pytest
    • Docker

  • AI application

    FrameGuard

    View repository

    An agentic, multimodal video-compliance pipeline that uses retrieval-augmented generation to turn transcripts and on-screen text into evidence-grounded violation reports.

    How it works
    • LangGraph workflow that indexes a video, then audits it against brand and compliance guidelines.
    • Azure Video Indexer extracts transcripts and on-screen text (OCR); Azure AI Search retrieves the relevant guideline passages.
    • GPT-4o produces a structured report with category, description, severity, and timestamps for each violation.
    • Evaluated retrieval quality, groundedness, latency, cost, and failure modes with LangSmith and Azure Application Insights.
    • FastAPI service, containerized with Docker, with CI/CD for automated testing and deployment.
    • Python
    • LangGraph
    • GPT-4o
    • Azure Video Indexer
    • Azure AI Search
    • LangSmith
    • FastAPI
    • Docker

  • AI application

    DocumentLoader — RAG Q&A

    View repository

    Ask questions about your own documents and get answers grounded in retrieved passages, with citations.

    How it works
    • Ingests PDFs, text files, and web pages; chunks and embeds them into a persisted Chroma store.
    • Retrieves top-k passages and answers with an xAI Grok model, streaming tokens and listing sources.
    • Three interfaces over one core package: CLI, Streamlit chat, and a FastAPI service with server-sent events.
    • Conversation memory across turns; Docker Compose for local runs.
    • Python
    • LangChain
    • Chroma
    • OpenAI embeddings
    • xAI Grok
    • FastAPI
    • Streamlit

  • AI application

    Corporate Brochure Generator

    View repository

    Turns a company website into a Markdown brochure using a language model at two steps.

    How it works
    • Collects links from a homepage, then asks the model which pages are relevant (About, Products, Careers).
    • Scrapes the selected pages and passes the combined content to the model to write the brochure.
    • Scraping utilities kept in a separate module from the notebook that runs the pipeline.
    • Python
    • xAI Grok
    • Requests
    • BeautifulSoup
    • Jupyter

  • Machine learning

    California Housing Price Prediction

    View repository

    A complete scikit-learn preprocessing and regression pipeline on the California Housing dataset.

    How it works
    • Stratified train/test split, imputation, one-hot encoding, and scaling inside one ColumnTransformer pipeline.
    • Ratio features, log transforms for heavy tails, and geographic similarity features from k-means cluster centers.
    • Linear regression baseline evaluated with test RMSE.
    • Python
    • scikit-learn
    • pandas
    • NumPy

  • Machine learning

    Machine Learning from Scratch

    View repository

    Regression algorithms implemented in NumPy with the math written out alongside the code.

    How it works
    • Linear regression by gradient descent and by the normal equation.
    • Locally weighted regression with a Gaussian kernel.
    • Python
    • NumPy
    • Jupyter

  • Data analysis

    U.S. Flight Delays and Cancellations

    View repository

    Exploratory analysis of roughly three million U.S. flights from 2019 to 2023 (public Kaggle dataset).

    How it works
    • Delay and cancellation patterns by airline, route, season, and time of day.
    • Breakdown of delay minutes by reported cause.
    • Time-based features (hour, weekday, season) and visual summaries.
    • Python
    • pandas
    • NumPy
    • Matplotlib
    • Seaborn

06 — Toolkit

Tools, listed where they were actually used.

  • Languages and data

    • Python
    • SQL
    • TypeScript
    • Polars
    • pandas
    • NumPy
    • DuckDB
    • SQL Server
    • Parquet
    • Jupyter

    Every professional study and most personal projects.

  • Modeling and statistics

    • scikit-learn
    • Regularized logistic regression
    • Naive Bayes
    • SciPy
    • Bootstrap intervals
    • Rank correlation
    • Agreement measures

    Credit-risk modeling, data-quality benchmarking, the public-data pipeline, and the housing model.

  • Record linkage and evaluation

    • Edit-distance similarity
    • n-gram similarity
    • Fellegi–Sunter linkage
    • Blind evaluation
    • Held-out and temporal validation

    Entity resolution and the public-data pipeline.

  • AI applications

    • LangChain
    • LangGraph
    • Retrieval-augmented generation
    • Tool calling
    • LLM evaluation
    • OpenAI / Anthropic APIs
    • Chroma
    • Azure OpenAI
    • Azure AI Search
    • Azure Video Indexer

    InsightDesk, FrameGuard, the RAG Q&A app, and the brochure generator.

  • Applications and tooling

    • FastAPI
    • Pydantic
    • SQLAlchemy
    • Streamlit
    • React
    • Next.js
    • pytest
    • Docker
    • Git and GitHub

    Product integration work, personal project interfaces, and this site.

07 — Contact

Have data that needs to be useful?

Open to 2027 Data Science, ML, and AI opportunities. The fastest way to reach me is email.

pranshughori203@gmail.com