The interactive CV of

Shrinidhi KJ

I'm Shrinidhi, an AI and ML engineer in Glasgow. I build LLM applications, retrieval systems and computer vision models, and I test that they work before I trust them.

Open to graduate and junior roles in the UK

About

Shrinidhi KJ in a dark suit and blue tie, standing outdoors

I'm an AI and machine learning engineer based in Glasgow, with hands-on experience building LLM and RAG systems, computer vision pipelines and deployed AI applications. I have just finished an MSc in Artificial Intelligence at the University of Stirling, after a B.Tech in Computer Science at PES University, Bengaluru.

My dissertation asked whether RAG systems ground their answers in retrieved evidence or recite what the model memorised: a diagnostic study over a 501-paper corpus with human-validated evaluation. I'm a published researcher, with internships at Nestlé and Zidio Development.

I work on LLM applications, retrieval-augmented generation, agents and computer vision. What I care about most is evaluation: a model that looks right is not always right for the right reason, so I build the checks alongside the system.

I use Claude Code every day, and I review and test what it writes before I rely on it.

I've just finished an MSc in Artificial Intelligence at the University of Stirling, after a B.Tech in Computer Science at PES University, Bengaluru.

My dissertation asked whether RAG systems really use what they retrieve, or just recite what the model memorised. I'm a published researcher, with internships at Nestlé and Zidio Development.

What I care about most is evaluation: a model that looks right is not always right for the right reason, so I build the checks alongside the system.

Undergraduate

Dec 2020 to Aug 2024

B.Tech Computer Science and Engineering

PES University, Bengaluru

Thesis: Generating Images from Text Descriptions using Diffusion Models and CLIP.

Published in GRADIVA Review Journal, May 2024.

Experience

Oct 2023 to Dec 2023

Data Analyst Intern

Nestlé, Bengaluru
- Collected, cleaned and explored data in Python (Pandas, NumPy), improving the defect resolution rate from 30% to 40%.
- Automated reporting workflows and dashboards, increasing reporting efficiency by 20%.
30% → 40%defect resolution rate

about20%faster reporting
Feb 2024 to Mar 2024

Data Science and Analytics Intern

Zidio Development, Bengaluru
- Analysed large, multi-source datasets to identify trends and actionable insights that supported data-driven decisions.
- Built and optimised Python data processing pipelines with Pandas, NumPy, SQL and R, reducing computation time by 30%.
- Translated analytical findings into business-relevant recommendations, working with people across teams.
about30%less computation time

Stirling, Scotland

Where the rain falls sideways and the coos don't care.

Postgraduate

Jan 2025 to Sept 2026

MSc Artificial Intelligence

University of Stirling, Scotland

Dissertation: Do RAG Systems Retrieve or Memorise?

Modules include

Machine Learning (Distinction)
Stochastic Processes and Optimisation (Distinction)
Deep Learning for Vision and NLP
Mathematical and Statistical Foundations

Graduation ceremony November 2026.

Selected work

What I built, how I checked it, and what the numbers actually show.

MSc dissertation, 2026

Do RAG systems retrieve or memorise?

When a retrieval system answers correctly, is it reading the documents or reciting what the model memorised? A five-condition experiment that changes only the documents the model sees.

Python
LangChain
ChromaDB
Llama 3.1 8B / 3.3 70B
DeBERTa NLI
Groq
GROBID

Built: 501 open-access papers parsed with GROBID into 34,502 chunks in ChromaDB with BGE embeddings. A five-condition experiment that varies only the supplied documents, to separate retrieval-grounded answers from parametric memory, across 120 cells with two Llama models (3.1 8B and 3.3 70B) through the Groq API.
How I checked it: The LLM-generated questions leaked their answers, so I rebuilt the evaluation set by hand and the conclusion reversed. I validated the LLM judge against two blind annotators (0.975 agreement, kappa 0.95), with an NLI model from a different family as an independent check.
Result: Retrieval drives correctness, not memory: accuracy rose from 33% with no documents to 92% with retrieval (McNemar p = 0.0001, replicated across both model sizes). Giving the model the exact source paper showed no detectable further gain (p = 0.5). Both models adopted a fluent counterfactual in 24 of 24 tests. Limits: 12 questions, one topic, one model family.

Correctness by document condition, pooled

A: no documents33%
B1: standard retrieval92%
B2: the exact source paper100%
C: irrelevant documents17%
D: contradictory documents0%

Giving the model the exact source paper showed no detectable further gain over standard retrieval (p = 0.5).

Applied research, 2026

Evidence report drafting pipeline

An LLM pipeline that drafts sections of an expert evidence report from scientific papers, with a person reviewing the output.

Python
LLM APIs
Retrieval
Claude Code

Private repository

Built: An LLM client with response caching, retryable versus fatal errors, and per-call token, latency and truncation logging, covered by a 50-check self-test. Built with Claude Code, reading and testing its output.
How I checked it: Froze a 38-finding gold set, with a recorded hash and a re-runnable regression check, before any generation ran.
Result: All 42 citations traced back to sources the model was given, but only 4 of 19 reachable findings were covered (11 counting partial matches). Well cited, but incomplete.

Open source, 2026

twin-api

Turns any OpenAPI 3.x spec into a running mock REST API, so integrations and AI agents can be tested without touching a live service.

Python 3.11
FastAPI
Pydantic
Faker
Prance
Docker

Code on GitHub

Built: Dynamic route registration, recursive $ref resolution, schema-valid synthetic responses, authentication from securitySchemes, and route ordering so parameterised paths don't hide specific ones.
How I checked it: Ran it against large real specs: Stripe (414 endpoints, 7.6 MB) and the GitHub REST API.

Computer vision, 2026

Brain tumour segmentation

Fine-tuned YOLOv8n-seg and YOLO26n-seg to outline brain tumours on MRI scans from BRISC 2025 (4,793 scans, 3 tumour classes).

PyTorch
YOLOv8
Gradio
Hugging Face Spaces

Try the demo on Hugging Face

Built: A training pipeline with a reproducible mask-to-polygon conversion, and an interactive demo on Hugging Face Spaces.
How I checked it: Compared both models on mask accuracy, parameter count and post-processing time.
Result: Mask mAP@50 of 0.905 and mAP@50-95 of 0.670. YOLO26n-seg used 18% fewer parameters and was 46% faster at post-processing.

B.Tech thesis, 2024

Text-to-image diffusion model

A text-to-image generation system using CLIP, a VAE, a UNet and diffusion models, trained on the Flickr30k dataset.

PyTorch
CLIP
Diffusion
AWS EC2

Built: Designed and trained the system, with training workflows deployed and scaled on AWS EC2. Published in GRADIVA Review Journal, May 2024.
How I checked it: Evaluated performance with FID and Inception Score.

Tools I work with

LLMs and RAG

Claude Code
Prompt and context engineering
Structured outputs
LangChain
ChromaDB
BGE embeddings
Vector databases
Hugging Face Transformers
Llama 3.1 / 3.3
Groq
DeBERTa NLI
GROBID
LLM-as-judge evaluation
Evaluation set design

ML and deep learning

PyTorch
TensorFlow
scikit-learn
YOLOv8
Computer vision
NLP
Supervised and unsupervised learning
Statistical validation

Programming and data

Python
SQL
R
JavaScript / TypeScript
Pandas
NumPy

Engineering

FastAPI
Pydantic
REST APIs
OpenAPI
Docker
Git
CI/CD
pytest
Linux
AWS EC2
GCP
Hugging Face Spaces
Gradio

Ready for my next quest.

I'm looking for graduate and junior AI / ML engineering roles in the UK. Email is the quickest way to reach me.

Send an email LinkedIn GitHub