h.kuchekar

●  BSc Data Science & AI · Leiden University

I turn raw data into things that work.

I'm Himanshu Kuchekar, a Data Science and AI student. I've built machine learning models, LLM pipelines and back-end services across healthcare, biology, sport and software, and I like owning a problem from the first messy dataset to the product someone actually uses.

See my experience Get in touch
4

Roles across AI, research and software engineering

6

Stages in an LLM pipeline I designed and shipped

413

Competencies covered by that pipeline's output

460+

Bacterial types identified in a sequencing study

About

Data first, domain second, people always.

I study Data Science and AI at Leiden University. What I enjoy most is the moment a pile of data becomes a model or a tool that changes how someone works.

That has taken me to some very different places: predicting Formula 1 race results, analyzing bacterial genomes, tuning back-end services, and building LLM systems for medical students. The domain changes. The method stays the same: understand the data, pick the right model, measure honestly, and ship something people can use.

I'm outgoing by nature, so I'm often the person connecting teams: running the stand-up, sitting with users, and turning what they say into requirements.

●  Experience · latest first

Four rooms, four kinds of data.

Medifjord.ai

AI/ML Engineering Intern

  1. Apr 2026 – Aug 2026

    Medifjord.ai

    AI/ML Engineering Intern

    Built the AI core of an education product, from raw textbooks to finished study tools, while acting as the link between three projects and the people they served.

    • Designed and shipped a six-stage LLM pipeline that turns textbooks into curriculum-aligned notes and flashcards covering 413 competencies across 82 topics.
    • Built the RAG corpus and OCR ingestion layer, and cut token spend with chapter pre-slicing, JSON templating and prompt caching.
    • Architected a four-model open-weight stack (Qwen 2.5, Hermes 3, Med42, DeepSeek-R1-Distill), evaluated through Hugging Face and benchmarked across Colab, Ollama and LM Studio for GPU-free testing.
    • Ran on-site user research and turned what users told me into product requirements.
    • Led daily stand-ups as the one common link across three concurrent projects, briefing the team on findings, new tools and competitors.
    LLM pipelinesRAGModel evaluationCost optimizationUser research

  2. Jan 2024 – Apr 2025

    Etchebarne Lab

    Undergraduate Research Assistant · East Lansing, MI

    Applied data science to a clinical bottleneck: getting from a biological sample to an infectious-disease diagnosis faster.

    • Analyzed genomic data with Python bioinformatics tooling to speed up infectious-disease diagnosis.
    • Contributed to predictive algorithms for rapid detection tests as part of an interdisciplinary team.
    • Helped build the software interface that makes the research usable in a clinical setting.
    PythonBioinformaticsPredictive modeling

  3. Oct 2024

    BMES Annual Meeting

    Presenter · Baltimore, MD

    Presented a sequencing analysis of freshwater from Niger, Nigeria and Burkina Faso to the Biomedical Engineering Society.

    • Identified more than 460 bacterial types from Nanopore sequencing data, including antibiotic-resistant E. coli and Salmonella enterica.
    • Collaborated with Université de Nantes to support water-quality monitoring in the region.
    Nanopore sequencingData analysisPublic speaking

  4. Jun 2024 – Aug 2024

    Acme Infovision

    Software Engineering Intern · Satara, India

    Worked on production back-end services, where the goal was making every request faster and more reliable.

    • Developed, debugged and profiled Python and FastAPI services.
    • Optimized request handling to improve responsiveness under load.
    PythonFastAPIProfiling

Selected work

Different problems, same toolkit.

Personal project · 2025

F1 Japanese Grand Prix predictor

An end-to-end ML pipeline that forecasts race outcomes from driver form, qualifying and circuit features. Random Forest and XGBoost were tuned with GridSearchCV and cross-validation and beat the baseline on accuracy and F1.

scikit-learnXGBoostpandasseaborn

Medifjord.ai · 2026

Textbook-to-flashcard LLM pipeline

Six stages take a scanned textbook through OCR, slicing and retrieval to structured notes and flashcards. Prompt caching and JSON templates keep the cost per chapter low.

RAGOCRPrompt caching

Medifjord.ai · 2026

Multi-model reasoning tutor

A four-model open-weight stack where each model handles what it's best at, from general reasoning to domain knowledge. It was benchmarked to run without a GPU budget.

Qwen 2.5DeepSeek-R1Ollama

Research · BMES 2024

Sequencing analysis of water samples

Processing Nanopore reads from three countries surfaced more than 460 bacterial types, including antibiotic-resistant strains, feeding a water-monitoring effort with Université de Nantes.

PythonNanoporeData analysis

Toolkit

What I work with.

ML & data science

  • PyTorch
  • TensorFlow
  • scikit-learn
  • XGBoost
  • pandas
  • NumPy
  • Feature engineering
  • Cross-validation
  • seaborn
  • matplotlib

LLMs & inference

  • Hugging Face
  • Transformers
  • RAG pipelines
  • Prompt engineering
  • Model evaluation
  • Quantization
  • Ollama
  • LM Studio
  • Token cost modeling

Software

  • Python
  • C++
  • TypeScript
  • JavaScript
  • FastAPI
  • Next.js
  • Node
  • Tailwind
  • Firebase
  • Git

Working with people

  • User research
  • Requirements
  • Stand-up facilitation
  • Cross-team coordination
  • Presenting
  • Competitor research

Contact

Have a role or a problem worth solving? Let's talk.

I'm open to internships, research positions and collaborations in data science, machine learning, AI and software engineering.

h.p.kuchekar@umail.leidenuniv.nl
Leiden, NetherlandsCET