Software Engineer, AI

I turn

ideas into
shipped software.

Full-stack LLM products, RAG workflows, and evaluation systems built with product sense, research depth, and a human-centered eye for the details that actually matter.

RAG & Retrieval LLM Evaluation Full-Stack AI Agentic Systems Knowledge Graphs UMass MS CS · 2026
🇺🇸 Work Auth OPT STEM OPT 🇺🇸 US E3 Visa · 🇦🇺 AU Citizen
See what I've shipped Let's talk
Priya Balakrishnan

Featured Work

Systems I've built
& shipped

✶ Shipped

01

AI/ML Engineer · Nevara AI

Full-Stack AI Recruiting Platform

End-to-end AI recruiting platform with written, voice, and role-play assessments. Spearheaded the Voice Assessment microservice, integrating Deepgram speech-to-text and GPT-4o into a 10-score evaluation rubric that cut manual review time by over 90%.

>90%Review time cut
~85%Match accuracy
100%Submissions processed
React / TypeScriptFastAPIAWS LambdaDeepgramGPT-4oLangChainSupabase
✦ Shipped

02

Full Stack Developer · Deloitte

Secure Enterprise Authentication Platform

Production-grade authentication system for Australia's largest FX provider, securing 100k+ customers. Integrated Auth0, MFA, and built automated C# migration tooling that moved 10k+ user records in under 4 minutes, reducing manual effort by 90%.

100k+Customers secured
60%Support tickets reduced
90%Manual effort saved
C# / ASP.NETReact / TypeScriptAuth0MySQLREST APIs
★ Research

01

Research Extern · Honda Research Institute

Git Intelligence Layer for Code QA

Repository-level code intelligence combining BM25 lexical retrieval with knowledge-graph traversal over function-level call graphs. Intent-aware pipeline using LLM-based query decomposition and a multi-metric evaluation harness across 1,700+ benchmark examples.

8–11%Top-1 localization lift
+13ptFile-level accuracy
1,700+Benchmark examples
LLM EvaluationBM25Knowledge GraphsDeepCodeBenchASTPython
✶ Research

02

Agentic AI Engineer · UMass CIIR

Multi-Agent RAG Evaluation System

LangGraph-based multi-agent RAG system for structured decomposition and iterative refinement of complex queries. Parallelized fact verification and coverage evaluation agents, with resource-aware execution for stable large-scale experimentation.

40–50%Execution time cut
100sBatch inputs evaluated
LangGraphMulti-Agent RAGParallelizationPythonLLM Evaluation

More projects & publication

Exploring the edges

Drag or scroll to explore →

Full-Stack AI Product

AutoCareers — AI Job Search Platform

Full-stack job search platform with intelligent resume parsing, ATS scoring, and cover letter generation. Multi-agent AI system achieving 95% parsing accuracy, reducing candidate job-hunt time by 45%.

React / TypeScriptFastAPIMongoDBPineconeAWSLlamaIndex

Applied NLP Research

Hierarchical Medical Note Generation

End-to-end hybrid NLP pipeline combining key phrase extraction, hierarchical clustering, and open-source models to generate clinically structured SOAP notes. 91% section accuracy, 15% factual grounding improvement.

NLPBART-MNLILDAHPC ClustersPython

Responsible AI

Fairness-Aware Facial Recognition

CNN-based fairness-aware model using the Seldonian Framework to mitigate racial bias in age prediction. Reduced racial bias by 20%, achieved 95% compliance with fairness constraints on FairFace dataset.

PyTorchSeldonian FrameworkCNNscikit-learn

LLM Research

Concept Learning under Differential Privacy in LLMs

Geometric and representational analysis of Gemma-2B and VaultGemma-1B using CKA metrics across 60 semantic categories. Differential privacy increased intra-class variance by ~90% and caused systematic depth misalignment.

Gemma-2BCKA AnalysisDifferential PrivacyPython

IEEE Publication

Visualising Physiological Cues in Individualised VR for Mental Health

Published at IEEE ISMAR, October 2023. Designed four VR visualisations for physiological data and conducted usability testing with 30+ participants. Literature review, design principles, and full research paper authorship.

IEEE ISMAR 2023VR / XRHCIMental HealthUsability Testing

Data Engineering

Stock Price Prediction with Parallel Processing

Stock price prediction model using PySpark MLlib, leveraging parallel computing and SQL to process large-scale financial data with runtime analysis comparing training and testing performance.

PySparkMLlibSQLParallel Computing
Scroll to explore

The journey

From Muscat to Amherst,
a path built across continents

📍 Muscat, Oman

The Beginning

Roots & Classical Dance

Grew up in Oman, spending over a decade training in Indian classical dance. Learning that every subtle gesture is a deliberate choice and a foundation in noticing what others overlook.

💃

Early
years

🎓

2018–
2023

📍 Brisbane, Australia

University of Queensland

BE (Hons) Electrical & Computer Engineering

Moved continents to pursue engineering. Discovered HCI, machine learning, and human-centered design. Published at IEEE ISMAR on VR for mental health.

📍 Brisbane, Australia

Full Stack Software Developer

Deloitte · Jul 2023 – May 2024

First industry role at a global firm. Engineered secure authentication for 100k+ customers and learned what it means to build software that real people depend on every day.

🏢

2023–
2024

🔬

2024–
2026

📍 Amherst, Massachusetts

MS Computer Science, AI & ML

University of Massachusetts Amherst · GPA 3.89

Crossed the Pacific to go deeper. Specializing in machine learning, NLP, and responsible AI. Where research meets practice.

📍 Amherst, Massachusetts

Agentic AI Engineer

UMass CIIR · Sept – Dec 2025

Designed multi-agent LangGraph RAG systems. Parallelized evaluation agents and built resource-aware execution for large-scale batch experimentation.

🧠

Sept–
Dec 2025

🚀

Jun–
Aug 2025

📍 New York City

AI/ML Engineer

Nevara AI · Jun – Aug 2025

Shipped a full-stack AI recruiting platform at startup pace. Built the Voice Assessment pipeline from zero, serverless, end-to-end, processing 100% of submissions.

📍 Amherst, Massachusetts

Graduate Research Extern

Honda Research Institute · Feb – May 2026

Repository-level code intelligence: retrieval, knowledge graphs, and benchmark-driven evaluation. Improved top-1 file localization by 8–11% across 1,700+ examples.

✦

Feb–
May 2026

?

Next

🌍 Anywhere great

The next chapter

Open to opportunities · May 2026

Looking for a team building impactful AI where I can own real systems end-to-end. If you're working on something meaningful, let's talk.

Capabilities

Technical stack

AI & LLM Systems

LLMsRAG SystemsLangGraphLangChainAgentic AILLM EvaluationPrompt EngineeringGPT-4o / OpenAIDeepgramBM25 RetrievalKnowledge GraphsHugging Face

Full-Stack Engineering

ReactTypeScriptPythonFastAPIC# / ASP.NETREST APIsSupabaseMySQLMongoDBPineconePyTorchscikit-learn

Cloud & Infrastructure

AWS LambdaAWS S3 / API GatewayDockerCI/CDGit / LinuxAuth0ServerlessJest / PytestSFT

The human behind the systems

About me

From training in Indian classical dance for a decade to writing code that powers global platforms, I've always been obsessed with the details that create real impact. Dance taught me how every tiny movement can change a story — computer science taught me the same about code.

Today, I'm a Master's student in AI & Machine Learning at UMass Amherst with hands-on experience across startups, research labs, and global companies. At Deloitte, I engineered secure authentication systems for 100k+ customers and automated data migration that slashed manual effort by 90%. At Nevara AI, I built an AI-powered recruiting platform that cut manual review time by 90% and boosted candidate-job matching accuracy to ~85%. My research at CIIR focuses on multi-agent retrieval systems that make large language models more consistent and reliable.

What inspires me? The idea that technology, when designed with human perspectives in mind, can reduce cognitive overload and augment how we live, work, and connect.

Technically, I'm fluent in Python, FastAPI, React/TypeScript, AWS, Hugging Face, and LangChain — but what sets me apart is my global journey across Oman, Australia, and the U.S., and the resilience I've built chasing opportunities across continents.

I'm not just interested in AI that works. I'm interested in AI that earns trust. My favorite kind of engineering problem is ambiguous at the start and measurable by the end. If you're building the future of AI and want a curious, impact-driven engineer on your team, let's connect.

Currently: wrangling agents, retrieval graphs, and edge cases. ✦

3Continents
10+Years of dance
3.89GPA
1IEEE publication

Let's build
useful AI
together.

If you're building the future of AI and want a curious, impact-driven engineer on your team, let's connect. Open to full-time roles in Software Engineering, AI, Full-Stack AI, and Applied AI. Available May 2026.

Actively seeking opportunities · May 2026
Emailbp42official@gmail.com LinkedInpriya-balakrishnan-sde GitHubgithub.com/Bpriya42 Phone413-512-3915