Ommkar Bisoi · Portfolio
Vol. 01 — Issue 2026 Bhubaneswar, IN Open to roles

Ommkar
Bisoi.

Building intelligent systems
at the intersection of language & the web.

A final-year Computer Science engineer at KIIT University, graduating in 2025. I work where machine-learning research meets shipped software — turning notebook prototypes into things people can actually use, with a recent focus on retrieval-augmented generation, transformer fine-tuning, and full-stack delivery.

See selected work Reach out

An engineer with a soft spot for language models, the messy parts of NLP pipelines, and shipping things end-to-end.

I'm reading my final year of B.Tech in Computer Science at KIIT University, finishing on an 8.13 CGPA. Most of my recent work sits at the boundary between research-grade ML and useful product — fine-tuning transformers, designing RAG pipelines, and wiring up the bits in between.

This past August I interned with Nalco, designing an automated technical-bid evaluation system using RAG and exploring on-premise deployment paths via Ollama and LlamaCPP for an internal-server environment.

8.13/10
CGPA · KIIT
23
Courses completed
3+
ML / NLP projects
1
Industry internship

Where I've spent real hours.

Aug 2024

Machine Learning Intern

National Aluminium Company (NALCO)

Designed a Retrieval-Augmented Generation system to score contractor submissions against required certifications and prior work history — turning a manual procurement check into an assistant that surfaces the right documents and rationales.

  • Translated ambiguous procurement criteria into structured retrieval prompts and scoring rubrics.
  • Evaluated on-premise deployment paths — Ollama and LlamaCPP — to fit Nalco's internal-server constraints.
  • Worked through the gap between "model returns plausible text" and "reviewer trusts the output enough to act on it."

A few things I've built.

Click any entry to expand it. Tools and tags are listed below each title.

01

LLM Summarization Web App

Fine-tuned T5-small on the BillSum dataset using the Hugging Face Trainer API, pushed the model to the HF Hub for pipeline-based inference, and shipped a Django backend that pipes a string through the model and returns the summary to the frontend.

Hugging FaceDjangoPythonTransformers
  • Configured the HF Trainer with a custom data collator and seq2seq training arguments — learning rate, weight decay, gradient accumulation tuned for limited GPU memory.
  • Pushed the fine-tuned checkpoint to the Hugging Face Hub so inference is one pipeline() call away — no local model files to ship.
  • Django view accepts a POST with raw text, runs it through the pipeline, and returns the summary as JSON to the frontend.
  • End-to-end this turned a research notebook into something a non-ML user can paste a bill into and read back a summary.
View on GitHub
Mar 2024
02

Sentiment Analysis · Naive Bayes

Built a Naive Bayes sentiment classifier on the Twitter 5000 dataset using NLTK — including tokenization, Porter stemming, and stopword removal in the preprocessing pipeline.

PythonNLTKNLPscikit-learn
  • Lowercased and tokenised tweets, dropped NLTK stopwords, then collapsed each surviving token to its Porter stem.
  • Built a frequency table of (token, label) pairs to drive a Multinomial Naive Bayes classifier from first principles.
  • Evaluated against a held-out split — the classic baseline that's surprisingly hard to beat on noisy short text.
  • Good warm-up for understanding why every modern NLP pipeline still uses these preprocessing primitives.
View on GitHub
Jan 2024
03

Movie Recommender System

Used the IMDB 5000 dataset to merge cast, director, and metadata into a single feature string per film, then computed a similarity index over those features to surface "if you liked X, try Y" recommendations.

PythonPandasContent-based
  • Joined cast, crew, genre, and keyword tables into a single per-film bag-of-tokens.
  • Computed a similarity score between films based on shared tokens — the simplest content-based recommender.
  • Surfaced top-N similar films for any seed title, with the score as a sanity check.
  • Useful baseline before reaching for collaborative filtering or learned embeddings.
View on GitHub
Oct 2023

Tools I reach for.

Click any item to scroll to and highlight projects that use it.

Languages

Python C++ C Java HTML & CSS SQL basic

Frameworks & Libraries

Hugging Face Django learning React.js learning NLTK Transformers LlamaCPP Ollama

Tools & Platforms

VS Code JupyterLab Jupyter Notebook PyCharm Spyder Eclipse Linux GitHub

Focus Areas

Machine Learning Deep Learning NLP RAG Systems LLM Fine-tuning Full-Stack Dev

What I've been studying.

Education

KIIT University
2021 — 2025
B.Tech, Computer Science & Engineering
CGPA · 8.13/10
St. Xavier's High School
2019 — 2021
Class XII · CBSE
81.4%
St. Xavier's High School
2009 — 2019
Class X · CBSE
85%

Certifications & Specializations

IBM Full-Stack Developer
12 courses
Professional Certificate · IBM via Coursera
Ongoing
Machine Learning Specialization
3 courses
deeplearning.ai · Coursera
Deep Learning Specialization
3 courses
deeplearning.ai · Coursera
NLP Specialization
5 courses
deeplearning.ai · Coursera

A note on leading people.

House Captain · Student Council

2017 — 2018
St. Xavier's High School

Oversaw house programs and coordinated committees, organising events that unified the houses while building strong camaraderie among members. Took ownership of planning, delegation, and on-the-day execution.

§ 07

Let's talk shop.

I'm exploring full-time roles in machine learning, NLP, and full-stack engineering. Always happy to talk about ML systems, RAG, or a good problem worth solving.

Email ommkar003@gmail.com Phone +91 97789 81952 LinkedIn Connect with me GitHub View repositories