dee.Contact me
Hey, I’m Dee

AI
Developer

Deekshitha Maganti

Build thoughtfully.
Evaluate rigorously.

I build AI systems and investigate how well their answers hold up to evidence.

Contact me
#01 LLM Evaluation#02 Grounded AI#03 Multimodal Systems#04 Azure & Python
Behind the work

Making AI outputs
easier to understand.
And verify.

I’m an AI developer with a background in computer science and information systems. I work on language-model evaluation, retrieval-augmented generation, and practical tools for inspecting AI outputs.

See selected projects
01 / EVALUATION

Ask better questions.

Compare models using structured rubrics for accuracy, citations, and reasoning.

02 / GROUNDING

Follow the evidence.

Connect generated claims to supporting passages in technical documentation.

03 / APPLICATION

Make it useful.

Turn model outputs into reviewable tools for faculty and domain-specific knowledge work.

How I approach AI systems

From a question
to an answer
you can inspect.

Across my projects, the work connects source material, retrieval, model outputs, and evaluation. The goal is to make errors visible and give people the context to judge an answer.

01 Structure questions & source material
02 Build retrieval & model workflows
03 Evaluate claims & failure patterns
04 Create reviewable summaries & tools
Selected work

AI applications &
evaluation projects.

A closer look at the systems I’ve built and the questions behind them.

Compare / Inspect / Evaluate
Questions
→
Models
→
Rubric
Workflow illustration

Engineering LLM Evaluation

Compared language models on engineering questions using structured rubrics for accuracy, grounding, citation quality, and reasoning.

Python · Model APIs · Streamlit · Power BI

More about the project

Created a labeled question dataset spanning hardware, networking, cloud, embedded systems, and cybersecurity. Built interactive views to compare models and review failure patterns.

Evidence before confidence
Claim
→
Source
→
Verdict
Workflow illustration

Technical Claim Verification

A retrieval and verification pipeline that checks model-generated claims against engineering documentation.

LangChain · FAISS · Azure Cognitive Search · Azure Functions

More about the project

Extracted factual assertions and retrieved supporting passages. Classified each claim as supported, unsupported, or contradicted, retaining evidence and confidence information for review.

See / Structure / Compare
Diagram
→
Vision AI
→
JSON
Workflow illustration

Engineering Diagram Interpretation

Compared vision models on circuit diagrams, network topologies, and cloud architecture diagrams.

GPT-4o · Claude Vision · Azure AI Vision · JSON Schema

More about the project

Structured model interpretations as JSON and compared them against ground-truth annotations. Reviewed component recall, description accuracy, and recurring interpretation errors.

Activity to insight
Logs
+
Feedback
→
Review
Workflow illustration

CEATS

A Claude-powered tool for university faculty to review individual contributions to group projects.

Python · FastAPI · Anthropic API · Azure SQL · Power BI

More about the project

Combined activity logs and peer feedback into contribution summaries with reasoning traces. Iterated prompts around classroom edge cases and documented the system for faculty use.

Knowledge with grounding
Question
→
Retrieval
→
Answer
Workflow illustration

AgriAdvisor

A RAG assistant for crop-management questions, grounded in agricultural extension documents.

Python · LangChain · FAISS · Streamlit · Azure App Service

More about the project

Added an evaluation harness for answer grounding, hallucinations, and task completion. Refined prompt templates through structured review as part of a computer science capstone.

Education

M.S. Information Systems

Murray State University · 2025–2026

B.Tech Computer Science

Lovely Professional University · 2020–2024

Get in touch

Let’s talk about
your idea.

Tell me what you have in mind.

This form uses FormSubmit to forward your name, email, purpose and optional phone number to my inbox. FormSubmit retains submissions for 30 days and uses Google reCAPTCHA. Privacy and terms. You can also email me at deemag.work@outlook.com.