About
AI Researcher with a PhD in Computer Science, working at the intersection of Explainable AI, natural language processing, and human-centred research design. Comfortable at every stage of the delivery lifecycle, from research-led prototyping through to production software: built AI-enabled systems for evidence retrieval, fact verification, and bilingual language learning, and shipped production-grade web platforms end to end with custom calculation engines and LLM-integrated features, alongside qualitative and survey-based research evaluating how communities engage with AI tools in education, health, and culturally responsive contexts.
Doctoral research developed an explainable, context-aware retrieval framework for automated fact verification, combining NLP, thematic modelling, and interpretable machine learning. Experienced translating technical AI research into accessible findings for academic, government, and community stakeholders, with particular depth in AI applications supporting te reo Māori and Indigenous-led research.
Experience
Product Engineer – AstroVallayil (Kerala Vedic Astrology Platform)
Jun 2026 – Present
Titan Tech Emirates Pvt Ltd
- Conceptualised, architected, and built AstroVallayil (astrovallayil.com) end to end as a live production service, from domain research, branding, and domain setup through to deployment and ongoing operation. Full-stack Next.js 14 (App Router, TypeScript) with a React/Tailwind front end and API routes.
- Engineered a custom rules engine, using an established scientific library only for the underlying astronomy while writing the rule logic from scratch as separated, focused modules (compatibility scoring, divisional charts, and dosha/yoga evaluation), each traceable to authoritative published sources.
- Built an LLM integration that works with either Gemini or OpenAI, grounding responses in the user's own computed data by injecting structured context tables into the prompt, with streaming output. Provider and model are selected at runtime via environment configuration, API keys are held server-side, and geocoding uses free OpenStreetMap Nominatim with custom autocomplete, eliminating third-party API spend entirely.
- Deployed and operated the production service on AWS Lightsail (Ubuntu 22.04, nginx, PM2 process supervision, TLS across multiple domains), with a one-command deploy script and documented deployment and recovery procedures.
Research Assistant – AI for Education and Community Innovation
Mar 2025 – May 2026
Auckland University of Technology
- Contributed to TechTahi, a research project exploring AI-enabled, low-barrier tools to strengthen digital capability and culturally responsive technology engagement within Māori and Pacific contexts.
- Worked across education, technology, and community perspectives to evaluate how AI tools are adopted and experienced by learners, informing inclusive research design.
- Contributed to research synthesis, technical reporting, and co-authored AI-focused publications arising from project activities.
- Led manuscript revisions and coordinated editor and publisher communications to see the project's paper through to publication in May 2026.
Doctoral Researcher – Explainable Retrieval-Augmented AI for Fact Verification
Mar 2020 – Aug 2025
Auckland University of Technology
- Designed, built, and evaluated CARAG, a context-aware retrieval-augmented framework that grounds fact-verification outputs in retrieved evidence while exposing local and global explanations of model decisions, published in Applied Sciences (2025) and presented at ACIS (2024).
- Extended CARAG with an unsupervised thematic context discovery method to advance macro-level explainability; the resulting paper was a Best Paper Award finalist at KDIR (2025).
- Authored a comprehensive technical review of explainability approaches in automated fact verification (Applied Sciences, 2023) and released framework code and an accompanying dataset on GitHub and Hugging Face to support reproducibility.
Research Officer – AI-Driven Language Learning Technology
Sep 2021 – Dec 2024
Auckland University of Technology
- Co-developed Te Kōhanga o te Tūī, an AI-enabled bilingual learning platform combining speech recognition and language technology to support early childhood acquisition of te reo Māori and English.
- Supported applied NLP research, technical documentation, and interdisciplinary collaboration across language, education, and AI-focused work.
- Contributed to stakeholder engagement and funding initiatives supporting AI-driven innovation and project development.
Research Assistant – AI Modelling for Tinnitus Detection
Mar 2022 – Mar 2023
Auckland University of Technology
- Contributed to interdisciplinary research using EEG, clinical, and psychological datasets to examine factors influencing tinnitus detection and treatment outcomes.
- Supported data preparation, feature analysis, and model interpretation within a multidisciplinary research environment.
- Developed analytical models and a minimum viable prototype (MVP) demonstrating practical application of AI-driven research findings.
Business Analyst
Oct 2017 – Mar 2019
Tureya Limited
- Led requirements gathering and stakeholder communication for mobile applications including user needs documentation.
- Delivered UX wireframes, data flow diagrams, and test plans supporting MVP delivery.
Earlier Industry Experience
Software Quality Assurance Engineer
May 2012 – Feb 2014
TechWyse Internet Marketing
- Built data-driven tests and integrated bug tracking with Mantis and SQL-based verification routines.
Jr. Software Quality Assurance Engineer
Nov 2010 – Apr 2012
ART Technology & Software India Pvt. Ltd
- Performed SQL-based backend validation, test case documentation, and system testing of e-commerce platforms linked with Microsoft RMS.
Selected Work
Kerala Vedic Astrology
AstroVallayil
Production web application built end to end: architecture, build, deployment, and operation. Next.js 14 (App Router, TypeScript) with a custom rules engine: an established scientific library handles only the underlying astronomy, while the rule layer itself is written from scratch as separated modules traceable to authoritative sources. The AI layer works with either Gemini or OpenAI, selected at runtime by configuration, grounding responses in the user's own computed data via context tables injected into the prompt, with no vendor lock-in and no third-party API spend. Deployed and operated on AWS (Ubuntu 22.04) with nginx, PM2 process supervision, and TLS, plus a one-command deploy pipeline and documented deployment and recovery procedures.
Explainable AI
CARAG — Explainable Fact Verification
Context-aware retrieval and explanation-generation framework for automated fact verification, using thematic embeddings and subset-of-interest graphs over FAISS. Released as framework code and a public dataset on GitHub and Hugging Face.
Māori Language AI
Te Whitinga Reo
Sovereign, culturally intelligent agentic AI for te reo Māori ↔ English communication, with TRANSLATE, INTERPRET, EXPLAIN, PRESERVE, and ESCALATE modes over Whisper ASR, a Māori-tuned LLM, and TTS.
Medical AI
Explainable Chest X-ray
MONAI DenseNet121 pneumonia classifier with Grad-CAM visual explanations and a Streamlit demo application, trained on Apple Silicon. Demonstrates interpretable diagnostic imaging with explanation overlays.
Data Engineering
AI Trip Planner
Databricks capstone: Spark ingestion pipelines, pgvector semantic search over destinations, live weather, and a FastMCP agent served from Databricks Apps, deployed as a live web app with the source published alongside it.
Adaptive Facade Simulation
ClimaFaçade
Interactive simulation of a smart adaptive building facade: a physics model for solar, glare, PMV, and HVAC paired with a TensorFlow.js surrogate ML model and performance dashboards for climate-responsive design exploration.
Health Data
EEG Tinnitus Research Tool
Electron and Flask desktop application processing live EEG streams over LSL, with feature engineering and machine-learning models supporting tinnitus research, packaged for use beyond the research setting. Repository private (participant data).
Data Pipelines
Weather Vector Search
Weather intelligence pipeline embedding NWS and Open-Meteo text with MiniLM, indexed for semantic search over pgvector with HNSW, and served through a Flask REST API and web interface.
Privacy-First AI
Rental Tax Assistant
Offline MVP analysing rental account statements with a local Mistral 7B via Ollama and LangChain. No data leaves the device; containerised and published on Docker Hub for any platform.
Agent Tooling
Weather MCP Server
FastMCP weather-prediction server (with umbrella reasoning) wired to a Databricks Agent Bricks agent, plus a Flask query dashboard for tool inspection.
Process Mining
Workflow Intelligence Assistant
Offline demo analysing Workday-style process exports: ranks manual steps by automation impact, answers free-text questions, and converts notes into structured pain points, entirely on a fully local model.
Publications
Journal Articles
Education Sciences · May 2026
Applied Sciences · Feb 2025
Applied Sciences · Nov 2023
Conference Papers
Books & Book Chapters
Book (Kindle & Print) · 2025 · Amazon
Datasets & Technical Reports
Dataset · Jan 2025
Technical Report · Dec 2022
Full publication list with abstracts: researchgate.net/profile/Manju-Vallayil-2
Education
PhD in Computer Science
Explainable AI & NLP — context-aware retrieval for automated fact verification
Auckland University of Technology (AUT) · School of Engineering, Computer and Mathematical Sciences
Mar 2020 – Dec 2025
Master of Information Technology
Eastern Institute of Technology
Mar 2016 – Apr 2019
Bachelor of Technology (BTech)
Computer Science & Engineering
Government Engineering College, Palakkad
Mar 2004 – Dec 2009
Skills
Explainable AI (XAI)
Natural Language Processing
Retrieval-Augmented Generation
Automated Fact Verification
Speech & Language Technology
AI for Indigenous Languages
Evidence Synthesis
Qualitative & Survey Research
Critical Thinking
Technical Writing
Stakeholder Communication
Research Ethics & Compliance
Python
LaTeX
Selected Methods
Context-Aware Retrieval (RAG)
Thematic Clustering & Embeddings
NLP Pipeline Development
Interpretable ML Modelling
Literature Review & Evidence Synthesis
Survey & Questionnaire Analysis
Awards & Certifications
Intelligent Replenishment Challenge — Salesforce (Apex invocable actions, Agentforce, Lightning carousel renderer for Consumer Goods Cloud)
2026
AUT Doctoral Journal Writing Grant
2025
AUT DCT Faculty Postgraduate Summer Research Award ×2
2023, 2024
Futures Pacific (CIIRID: IDEA) – AUT-PUCV Chile Scholarship ×2
2021, 2023
AUT ECMS Research Scholarship
2022
PhD Full Tuition Fee Waiver
2021–2024