Oltan Sevinc — Perception & ML Engineer

Experience

Founder & ML Engineer · Hango

Jul 2025 – Present

An event recommendation and social coordination app for Sydney, live on the App Store and Google Play. I built and run the entire stack as the only engineer.

Hango's personalized feed of Sydney events. — These screenshots are seeded with demo data. The live app ranks against a Postgres database of around 11,000 event templates across 2,000 Sydney venues.

Hango's map view, with events pinned across the Sydney harbour foreshore. — These screenshots are seeded with demo data. The live app ranks against a Postgres database of around 11,000 event templates across 2,000 Sydney venues.

Recommender. Personalized feed written in PL/pgSQL over pgvector. Each person is modelled as a handful of weighted interest centroids, updated as they use the app. The ranking spends some of its slots on safe matches and the rest on deliberate reaches. Written up here.
Group scoring. Recommending to several people at once with a max-plus rule, so one strongly matched member can carry the choice instead of the group averaging into something nobody wanted.
Data pipeline. An LLM ingest pipeline that scrapes unstructured listings and resolves them against what is already stored. It has turned about 90,000 raw events into 11,000 clean templates across 2,000 venues, with human review held under 1%.
Serving and infrastructure. A Deno and TypeScript edge-function API covering search, the group itinerary builder, push notifications and place resolution. Row-level security on every table, instrumented with PostHog and Sentry.
Frontend. The cross-platform app in React Native and Expo, on iOS, Android and a shareable webview.

Teaching Assistant · UNSW

Sydney

Feb 2022 – Present

Tutorials and lab sessions in robotics, state estimation and postgraduate AI.

Robotics (MTRN4230). Forward and inverse kinematics, DH parameters and trajectory planning.
State estimation (MTRN4010). Extended Kalman Filters and sensor fusion.
Postgraduate AI (COMP9414). Search, machine learning and reasoning.

Software Engineer · Honeywell

Intern, retained part-time · Sydney

Dec 2021 – Sep 2022

Joined as a summer intern and retained part-time through the academic year on the backend of Experion, Honeywell's flagship process-control platform.

Developed backend features in modern C++ with Boost, and automated the nightly build-archiving process in Python.
Worked in an Agile team, using JIRA, Confluence and Git.

Publications & research

Published · ACRA 2025

Towards Closing the Domain Gap with Event Cameras

M. Oltan Sevinc, Liao Wu, Francisco Cruz

Australasian Conference on Robotics and Automation · first author

Compares event cameras with grayscale frames for end-to-end driving across day and night. Models trained on event data degrade much less under lighting shifts, since event cameras respond to relative brightness change rather than absolute intensity.

Conference

Event-camera frame of a road beside a hillside in daylight: rock texture, road markings and the horizon picked out against flat grey. — Each sensor pair was captured simultaneously; day and night are comparable stretches of road, not the same one. The event stream changes little between them; the frame camera collapses into glare.

Grayscale frame-camera image of the same daylight scene, captured simultaneously: hillside, road and lane markings all legible. — Each sensor pair was captured simultaneously; day and night are comparable stretches of road, not the same one. The event stream changes little between them; the frame camera collapses into glare.

Under review

Continuous Attribution on Discrete Spikes: Revealing the Explainability Paradox in Spiking Neural Networks

M. Oltan Sevinc, Liao Wu, Francisco Cruz

Neuromorphic Computing and Engineering (IOP) · first author

Class activation maps explain conventional networks well and break in specific ways on spiking ones, where the signal is discrete and spread over time. The paper works out why, shows that the usual way of smoothing an explanation pushes sparse event data somewhere no sensor could have taken it, and proposes a noise model that stays inside what the data could plausibly be. It closes with a practitioner’s guide to which method to reach for, on which kind of input.

An event-camera frame of a person performing a gesture, accumulated over time: red events on black. — One DVSGesture sample. Gaussian noise pushes the input somewhere no sensor could have taken it, and the explanation slides off the subject onto empty background; Poisson noise keeps it on the person.

Standard GradCAM heat map over the same frame, with the hot region to the left of the subject. — One DVSGesture sample. Gaussian noise pushes the input somewhere no sensor could have taken it, and the explanation slides off the subject onto empty background; Poisson noise keeps it on the person.

Honours Thesis · 2023

Robotic Teleoperation with Haptic Feedback for Remote Ultrasounds

B.E. Mechatronic Engineering (Honours) · supervised by Liao Wu

UNSW Sydney

A real-time haptic teleoperation interface between a Universal Robots UR5e arm and a 3D Systems Touch device over ROS and MoveIt, using quaternion-derived angular velocity, deadband and consecutive-zero filtering, and force feedback from the arm's own sensor.

A hand holding the stylus of a 3D Systems Touch haptic device on a workbench, with a Universal Robots UR5e arm posed behind it and the teach pendant in front. — A 3D Systems Touch haptic device driving a Universal Robots UR5e over ROS and MoveIt. Demo video →

Education

PhD, Computer Science · UNSW Sydney

Feb 2024 – Present

Thesis: Applications of Spiking Neural Networks in Robotics.
Supervised by Liao Wu and Francisco Cruz. Australian Government Research Training Program (RTP) Scholarship.

B.E. Mechatronic Engineering (Honours) and Computer Science (AI) · UNSW Sydney

2018 – 2023

Honours thesis on robotic teleoperation with haptic feedback for remote ultrasound, supervised by Liao Wu.

Technical skills

Languages

C++PythonSQL · PL/pgSQLTypeScript / JSMATLAB

Robotics & Perception

ROS / ROS2MoveItGazeboExtended Kalman Filterssensor fusionstate estimationevent-based visionOpenCVLIDARkinematics

ML & Data

PyTorch · DDPSNNs · SpikingJellyrecommender systemsvector search / pgvectorLLMs & LLM ETLscikit-learn

Systems & Infra

PostgreSQL / pgvectorSupabaseDeno edge functionsETL pipelinesGCPDockerGitLinuxPostHogSentry

Writing

Histogram of pairwise cosine distances between event embeddings, peaking near 0.16, with the 0.09 cluster-match threshold marked to the left of the mass.

Recommenders18 Aug 2026

How to build a recommender system in Postgres

The decisions behind a personalized feed on Postgres and pgvector: representing taste, choosing thresholds, candidate generation, popularity with few users, and cold start.

Pipeline diagram: a dozen feeds and ninety thousand raw rows normalized onto one schema, resolved against what is stored, then merged or kept as new, leaving eleven thousand templates.

Data pipelinesNext

How to deduplicate scraped listings with an LLM

Normalizing records from a dozen sources that disagree, resolving them against what is already stored, and deciding which near-duplicates are the same event. Hango's ingest as the worked example.

Two night-time frames of the same road side by side: the frame camera washed out by glare, the event camera still resolving the lane and reflector posts.

PerceptionLater

How to keep a vision model working after dark

Why a model trained on event-camera data barely notices when the lights go out, what that costs elsewhere, and when the swap is worth making. From an ACRA 2025 paper on the day-night domain gap.

→ Read the blog