AI · 2026
Live<30s
HireIQ — AI Candidate Intelligence Platform
CV + transcript in, client-ready candidate report out — under 30 seconds. Manual baseline was 30–45 minutes.
Python · FastAPI · Claude API · React · Railway · Vercel · PostgreSQL
AI Engineer · Apprento.io · Auckland
Roshan Aryal
I find the hidden boundaries in deployed AI systems — runtime ceilings, inferential limits, and the decisions that determine who they actually serve.
Currently building HireIQ at Apprento.io while finishing my Bachelor of IT at Otago Polytechnic Auckland.
- HireIQlive at Apprento.io
- 88.0% R²Random Forest
- 5 projectsall shipped
Selected work
Each one shipped. Each one written up.
+1ML · 2025
Live88.0% R²
Road Accident Risk Predictor
Random Forest on real crash data, behind a Streamlit UI a non-technical user can drive.
Python · scikit-learn · Random Forest · Streamlit
+2ML · 2025
AUC 0.81
Diabetes Prediction Model
KNN-imputed Pima dataset, threshold-tuned for recall over accuracy. ROC AUC 0.81.
Python · scikit-learn · pandas · GridSearchCV
Web · 2024–25
Live
Quiz Tournament Platform
Three-role Spring Boot platform. Live tournament state via versioned polling — no websockets needed.
React · Spring Boot · PostgreSQL · JWT
Web · 2024
Live
RentMate — Equipment Marketplace
Three-role equipment marketplace. Live availability via Firestore listeners — no polling, no double-booking.
React · Firebase · Firestore · Tailwind
Research & manuscripts
Three manuscripts currently under peer review.
Manuscript under review · 2026
Runtime limits in browser-native AI retrieval systems
Identifying where in-browser semantic search silently breaks as data scales, and what structural fix removes the failure mode entirely.
Manuscript under review · 2026
Causal inference in observational road-safety data
Testing whether severity-based risk inference holds up once exposure and reporting-completeness confounds are accounted for.
Manuscript under review · 2026
Consensus and failure modes when LLMs design the same system
Eleven frontier models given one identical hard design task under a fixed hardware budget: where they converge, where they diverge, and which failure modes are measurable without an answer key.
Full titles, DOIs, and code will be linked here once each review concludes.
About
I’m an AI engineer at Apprento.io and a final-year Bachelor of Information Technology student at Otago Polytechnic Auckland (graduating December 2026). I prefer work that ships — every project below is live, with real users or real evaluation data behind it.
Most recent is HireIQ, the AI candidate-intelligence platform I’m building at Apprento — Claude-powered assessment reports that took recruiters 30–45 minutes manually, now generated in under 30 seconds with a human-in-the-loop review step. Before that, the work I keep coming back to is applied ML — the Road Accident Risk Predictor at 88.0% R² was the project that made it stick.
On the engineering side I work end-to-end — Python / FastAPI for the AI layer, React / Next.js on the front, Spring Boot or Node on the back, Firebase or Postgres for state. I lean toward fewer abstractions, fewer dependencies, and shipping the thing.
Currently — building HireIQ at Apprento.io alongside finishing my degree.
Skills
Reach for first
Python · FastAPI · Claude API (Anthropic) · scikit-learn · pandas · React · Next.js · PostgreSQL · Firebase
Comfortable in
JavaScript · Java · SQL · Node.js · Spring Boot · Tailwind CSS · Streamlit · Docker · AWS · Git
Currently exploring
LLM evals & guardrails · Agent tooling · TensorFlow time-series
Experience
Work
AI Engineer · Apprento.io
Apr 2026 – presentBuilding HireIQ, an AI candidate-intelligence platform that takes a CV plus an interview transcript and produces a client-ready assessment report in under 30 seconds — down from a 30–45 minute manual baseline. Stack: Python / FastAPI, Anthropic Claude API, React, Docker, AWS.
Education
Bachelor of Information Technology
Otago Polytechnic — Auckland International Campus
2024 – Dec 2026 (in progress)
Machine learning, software development, cloud computing
Timeline
- 2024
Started Bachelor of IT
Otago Polytechnic Auckland. Picked the ML / software-development / cloud track.
- 2025
Year of shipping
Four projects from idea to live URL — two ML, two web. The set that’s in the projects above.
- Apr 2026
AI Engineer at Apprento.io
Joined as AI Engineer at Apprento.io, building HireIQ — the candidate-intelligence platform. Moved from coursework into shipped product work.
- Aug 2026
First research manuscript submitted for peer review
Submitted a manuscript on runtime limits in browser-native AI retrieval to peer review. Currently under double-blind review.
- Dec 2026
Expected graduation
Open to full-time AI / ML / full-stack engineering roles in NZ and remote.
Notes
Short technical writeups from real project work.
Contact
I’m open to graduate roles, contract work, and a good conversation. The fastest path is email.