Ayan Ali Saleem

AI/ML Engineer Specializing in 3D Perception, Robotics, and LLM Reliability

Based in: Saint Petersburg, RussiaMost recently: Lab Engineer (Lab Demonstrator), ITMO UniversityExperience: 2 years

MSc Robotics & AI engineer with a Red Diploma from ITMO University and a background in electrical engineering. Expert in building and evaluating machine-learning systems for 3D reconstruction, LiDAR-camera fusion, and multi-agent LLM frameworks with a focus on reproducible, metrics-driven deployment.

Experience

Lab Engineer (Lab Demonstrator)

ITMO UniversitySaint Petersburg, RussiaworkDec 2024 – Dec 2025

Summary

Served as a Lab Engineer at ITMO University for approximately one academic year while completing an MSc in Robotics & AI, supporting Bachelor's-level engineering coursework. Taught and mentored undergraduate (Bachelor's) students directly, helping bridge lecture material with hands-on application. Conducted lab demonstrations, guiding students step-by-step through practical exercises and troubleshooting as issues came up.

Results

Supported Bachelor's-level coursework for about 25 students a week across 3 different lab groups.

Operations Management Engineer

Sapphire Renewables (Wind Power Company Ltd.)Karachi, PakistanworkFeb 2024 – Sep 2024

Summary

Assessed wind turbine data for predictive maintenance, improving uptime. Scheduled and supervised maintenance to minimize energy losses. Produced detailed breakdown reports and SOPs for plant operations. Conducted performance analysis to enhance turbine energy output.

What I did

Assessed turbine data to identify predictive-maintenance needs and scheduled maintenance around NPCC dispatch instructions to preserve dispatchable hours.
Supervised maintenance activities based on turbine condition and grid requirements.
Tracked operational performance through detailed breakdown reports and SOPs.

Results

Improved uptime through predictive maintenance assessment.

Performance analysis

Trainee (R&D)

Winston Marriot Ltd.Karachi, PakistaninternshipDec 2023 – Feb 2024

Network and firewall configuration within an R&D engineering function.

Skills

Technical

Python

Model Benchmarking

Object Detection

YOLO11

PyTorch

Git/GitHub

YOLOv8

YOLO12

Streamlit

Instance Segmentation

NeRF

LSTM

ARIMAX

LiDAR-camera sensor fusion

KITTI calibration

open3d

LangChain

LangGraph

Multi-agent systems

RAG

Hugging Face

ROS

MuJoCo

DWA

MPC

MATLAB/Simulink

Linux

FastAPI

PID

Volumetric Rendering

DBSCAN

Chroma

Stereo Vision

Gazebo

Shapely

UKF State Estimation

SLAM

PCA

C++

Model Context Protocol (MCP)

Pydantic

JSON Schema

Sequential Tool Orchestration

FAISS

Tavily

create_agent

Structured Outputs

Languages

English

Projects

MSc Thesis: Real-Time 3D Perception for Autonomous Ground Vehicles

Lead Researcheracademic

Training-free pipeline: YOLOv8m-seg → calibrated LiDAR projection → adaptive mask erosion → DBSCAN clustering → PCA 3D OBB fitting. Results: 50.51% Car IoU, 38.63% Pedestrian IoU on KITTI; 1,874 MiB GPU memory; ~11 FPS; 6.4× more memory-efficient than TransFusion (12,000+ MiB). Ablation study over 7,481 KITTI frames.

PythonPyTorchYOLOv8open3dShapelyKITTI calibration SDK

Citation Reliability of Small Open-Weight LLMs

Researcheracademic

Built a two-model evaluation (Phi-3-mini 3.8B, humains-junior 3.8B) of reference hallucination across 45 topics in three familiarity conditions (famous / lab-specific / post-cutoff). Verified 400+ model-generated references against Crossref, OpenAlex and arXiv; classified each as REAL_OK, REAL_BAD_LINK, WRONG_PAPER_LINK or FABRICATED. Found that even famous-topic requests produced 61% non-verifiable references for the best model, while niche and recent topics collapsed to ~95% fabricated.

Phi-3-minihumains-juniorCrossref APIOpenAlex APIarXiv API

Object Detection for Autonomous Vehicle Navigation using YOLO

Developeracademic

Designed and implemented a complete object detection pipeline using YOLOv8, YOLO11, and YOLO12 on the KITTI autonomous driving dataset. Conducted systematic comparative evaluation across architectures using mAP@0.5, mAP@0.5:0.95, precision-recall, FPS, FLOPs, and model size.

YOLOv8YOLO11YOLO12KITTI dataset

BE Thesis: Hybrid PV System Design & Smart Automation

Lead Designeracademic

Designed a hybrid photovoltaic system addressing Constant Power Load (CPL) instability. Components: Single-Phase Hybrid Inverter, Phase-Locked Loop (PLL), MPPT algorithm, CPL system. Methodology: theoretical analysis + Simulink MATLAB simulation; precise circuitry modelling.

MATLABSimulinkMPPTPLL

Tri-Nav: Integrated A*, DWA, and MPC for Robot Navigation

Developeracademic

Developed a hybrid navigation framework for ground robots in a MuJoCo simulation. Integrated A* (global), Dynamic Window Approach (local), and MPC (control). Focused on robustness in cluttered maze environments and real-time execution.

MuJoCoA*DWAMPC

View the code

Neural Radiance Fields (NeRF) for Scene Reconstruction

Developeracademic

Implemented NeRF pipelines for continuous 3D scene representation and novel view synthesis. Worked with multi-view image datasets, camera pose estimation, and volumetric rendering. Analyzed reconstruction quality, convergence behavior, and data requirements. Applied NeRF concepts to imaging-based 3D reconstruction relevant to robotics perception and mapping.

NeRFVolumetric RenderingCamera Pose Estimation

Multi-Agent Study & Coding Assistant with Hallucination-Detecting Reviewer Agent

Developeracademic

Built a LangChain/LangGraph multi-agent system (router, supervisor, researcher, writer, critiquer, code helper, quiz helper and Reviewer) with Chroma-based RAG, FastAPI backend and Streamlit UI. Designed a lightweight Reviewer Agent that scores code outputs on correctness, edge cases, security, quality and relevance, and triggers a refined loop below a threshold; evaluated on SWE-bench Lite in three conditions (baseline, Reviewer + refine, OpenHands agent under Reviewer supervision). Ran six 30-instance paired A/B cells across generator and judge families. Best cell (Mistral 7B, qwen3-32b judge): judge score 0.511 to 0.710 (+0.199, t = 3.85, p < 0.001).

LangChainLangGraphChromaRAGFastAPIStreamlitMistral 7Bqwen3-32bSWE-bench LiteTavilycreate_agent

applied-llm-systems

personal

Built a system to extract and validate data from messy invoice and resume text into structured Pydantic objects.

Pydantic v2JSONPython

Education

MSC Robotics and Artificial Intelligence

ITMO UniversityRobotics and Artificial IntelligenceGPA 4.852024 – 2026

Thesis. Real-Time 3D Perception for Autonomous Ground Vehicles

BE Electrical Engineering

NED UNIVERSITY OF ENGINEERING AND TECHNOLOGYElectrical EngineeringGPA 3.42019 – 2023

Thesis. Hybrid PV System Design & Smart Automation

Recognition

Awards

Participation Certificate - Global HackAtom

ROSATOM2025

Contributed to the development of innovative nuclear-technology solutions during World Atomic Week.

Winner - 2nd Round of HackAtom 2025

ROSATOM2025

Presented 'NukeNet - A Gamified Knowledge & Networking Hub' to connect flagship universities.

2nd Place - ObninskTech Winter School Competition

ROSATOM2025

Led Group 3.1 (The Co-Creators), investigating hydrophobicity of low-friction surfaces in steam turbines.

Winner - Hackathon 2025 (Rosatom Student Competition)

ROSATOM2025

Developed Nuke Net - A Gamified Knowledge & Networking Hub.

Certifications

Deep Learning Specialization

Coursera

Advanced Data Analytics

Google

AI Engineering

IBM

IoT Programming Specialization

Coursera

Data Science & Excel Foundations

Coursera

Patents & publications

TriLayerNav: A Tri-Layer Navigation System for Unmanned Ground Vehicles Integrating A*, DWA and Model Predictive Control

publication2026

Published in Gyroscope and Navigates (Scientific and Technical Journal of IT, Mechanics and Optics).

With Ali A.

Read it

Improving NeRF Training Quality Using Depth Data: A Comparative Study of Depth-Integration Strategies for 3D Scene Reconstruction

publication2026

Published in International Journal of Computer Vision (IJCV).

With Ali A.

Read it

A Hybrid ARIMAX-LSTM Framework for Robot-Assisted PM2.5 Forecasting and Adaptive Environmental Monitoring

research paper2026

Under Review at Atmospheric Research. Manuscript Number: ATMOSRES-D-26-03034.

With Ali A.

Social impact & activities

IT Team Member

NED IEEE SocietyvolunteerMay 2022 – Aug 2023

Developed a simple native JS DOM Manipulation Templating Library for the IEEE blog.

ObninskTech Summer School

ObninskTech Summer SchoolworkshopJun 2025

Engaged in seminars on AI, nuclear, and digital technologies, fostering innovation and collaboration.

ObninskTech Winter School

ObninskTech Winter SchoolworkshopFeb 2025

Participated in a prestigious international nuclear science program organized by Rosatom, MEPhI, and ObninskTech, collaborating with students and professionals from 39 countries.

Hire AyanReach out about a role, a contract or a conversation.For recruiters

Ayan's twin is AI, it can make mistakes.

More people on Kanvis

Browse more people Create your own Kanvis