200k
token memory & live step flow
CLOUD & AI ENGINEER AT GOOGLE SINGAPORE
Elias Lim
Building secure cloud platforms, backend services, reliable machine learning models, and practical AI developer tools.
CLOUD RUN, GEMINI PLATFORM & AGENT TOOLING
Google Singapore
Customer Engineer at Google Singapore architecting secure Cloud Run and Gemini platform systems, creator of Antigravity Agent Tracer (1st Place, Google AI Agent Lab Hackathon), and technical lead for buildathons training 4,000+ engineers across GovTech, DBS, NTU, and A*STAR.
EXPERIENCE & EDUCATION
Career progression. Cloud, systems, and research.
Engineering roles across cloud infrastructure, high-scale backend systems, and applied machine learning research.
Customer Engineer at Google
Architecting secure cloud and agentic AI workflows on Cloud Run, the Gemini platform, and Terraform. Built the native Antigravity Agent Tracer execution visualiser and two-by-two Antigravity Harness workspace with live 200k context telemetry (awarded 1st Place at Google's AI Agent Lab Hackathon 2026), and led AI buildathons training over 4,000 engineers and students across NTU, GovTech, DBS Bank, and A*STAR.
Trust Software Engineer at Grab
Built scalable Golang backend services for real-time fraud detection, implementing runtime control mechanisms such as feature flags and rate limiting on petabyte-scale risk infrastructure. Expanded automated unit test coverage, resolved production defects, and contributed to architecture and code reviews across Agile sprints.
Sep 2024 to May 2025
A*STAR Singapore
AI Research Intern at A*STAR (Agency for Science, Technology and Research)
Designed and implemented a unified uncertainty-aware deep learning framework for battery State-of-Health estimation that quantifies both model and data uncertainty alongside model-agnostic explainability (SHAP and LIME). First author of "A Unified Framework for Interpretable and Uncertainty-Aware Battery State of Health Estimation Using Deep Neural Networks" (APSIPA ASC 2025 / IEEE Xplore).
BSc (Hons) Computer Science (Second Upper Class), University of Glasgow & SIT
Awarded 2nd Place Overall at NUS, Singtel, and Millennium Management LifeHack 2025 (EduVerse), Overall Best (Intermediate) and Most Impactful at the AI Singapore, SMU, and CyberYouthSG Hackathon, and Semifinalist at DSTA BrainHack 2024 TIL-AI (Autonomous Air Defence System).
HACKATHON HONOURS
4x Award Winner / Finalist
Competitive AI & Systems Engineering
BY THE NUMBERS
By the numbers. Built for production.
Compare the automated AI systems against traditional manual baselines, or click any tile below to explore the repositories.
KEY HIGHLIGHTS
Get the highlights. Every signal at a glance.
95.2%
verified confidence range
AI models that quantify confidence and explain every driver.
4,000+
engineers & students trained
Hands-on cloud and AI buildathons across Singapore's top institutions.
1ST PLACE, GOOGLE AI AGENT LAB HACKATHON 2026
Antigravity Agent Tracer & Harness
Awarded 1st Place at Google's AI Agent Lab Hackathon 2026. Watch every step an AI agent takes live, from user prompt and planning to tool calls and subagents, inside a unified workspace with a 200k token memory meter.
LIVE AGENT STEP VISUALISER (CLICK ANY STEP)
Step 02: AI Planner (640ms)
STEP 1 OF 4: HUMAN GOAL
Click any box above to see what it means
A person sets the goal
Everything begins here. Instead of writing technical instructions, a person simply asks the AI to check a payment file so customers are never double-charged.
Turns raw background logs into an interactive visual timeline so engineers can inspect every prompt, reasoning step, and tool call at a glance.
AUTOMATED EXECUTIVE BRIEFINGS & FACT CHECKER
Smart Meeting Prep & Dossier Agent
Pulls together emails, chats, calendar invites, and documents into clean one-page meeting briefings, backed by 36 automated quality checks so links are never made up.
STEP 1 OF 3: 5 PM EVENING RUN
Click any column above to inspect
Spots real external & cross-team calls automatically
Every weekday at 5 PM, the assistant looks ahead at tomorrow's calendar, skips team stand-ups, focus blocks, and declined invites, and researches only the meetings that need preparation.
Explore three real calendar schedules right inside the browser, inspect generated one-page briefings, and test the live fake-link blocker.
FIRST AUTHOR, APSIPA ASC 2025 & IEEE XPLORE (DOC 11249263)
Uncertainty & Explainable AI for Battery Health
Deep learning framework that predicts battery health alongside a guaranteed confidence range (95.2% accuracy) and explains which sensor signals drove every forecast.
STEP 1 OF 3: RAW TELEMETRY INGESTION
Click any stage above to inspect
Multi-sensor discharge curves capture subtle ageing shifts
Streams terminal voltage drop (dV/dt), internal resistance growth, and surface cell temperature across thousands of McMaster and Oxford cycling curves.
Interactive digital twin plotting CNN State-of-Health forecasts alongside calibrated Adaptive Conformal Inference safety bands and 4-method XAI sensor attributions.
ALL REPOSITORIES
Explore the line-up. Built end to end.
Filter by engineering domain or expand any project tile to inspect its technical architecture.
SIDE-BY-SIDE ARCHITECTURE COMPARISON
Compare systems. Side by side.
Select any two repositories to compare their runtime stack, verification gates, and system schematics.
PUBLICATIONS
Peer-reviewed research. Published in IEEE Xplore.
Published in IEEE Xplore, with one-click BibTeX citations.
A Unified Framework for Interpretable and Uncertainty-Aware Battery State of Health Estimation Using Deep Neural Networks
Designed a deep learning framework that wraps every battery lifespan prediction in a verified 95.2% confidence range (with 0.012 calibration accuracy) and explains which physical signals, such as voltage and heat, drove each forecast across the McMaster and Oxford datasets.
EMPIRICAL COVERAGE
95.2% PICP
Target >= 95.0% across ageing cycles
CALIBRATION ERROR
0.012 ECE
Distribution-free Adaptive Conformal
BENCHMARK DATASETS
McMaster & Oxford
Dynamic drive-cycle & constant-current cells
ATTRIBUTION SUITE
4 XAI Methods
SHAP, LIME, Integrated Gradients & CF
Abstract: Accurate State-of-Health (SOH) estimation is critical for safe lithium-ion battery operation, yet point-estimate deep neural networks provide neither calibrated confidence bounds under distribution shift nor physical interpretability. This paper introduces a unified framework pairing convolutional SOH estimation with Adaptive Conformal Inference (ACI) to guarantee finite-sample coverage across non-exchangeable ageing trajectories, alongside cross-validated SHAP, LIME, Integrated Gradients, and Counterfactual explanations that isolate voltage relaxation, incremental capacity peaks, and thermal rise at knee-point degradation.
Teamwork Assessment in Software Engineering Education
Evaluated practical data-driven methods for measuring collaboration and engineering contribution patterns across software student teams.
TELEMETRY SIGNALS
Git, PR & Peer Data
Commit cadence, code review & sprint velocity
EVALUATION SCOPE
Multi-Team Cohorts
Longitudinal software engineering capstones
ASSESSMENT OUTCOME
Fair Contribution
Reduces subjective peer-grading bias
PUBLICATION VENUE
IEEE Xplore (2025)
Document 10892827
Abstract: Assessing individual contributions within collaborative software engineering projects remains challenging when relying solely on end-of-term peer surveys. This study investigates combining repository activity, pull-request review interactions, and structured milestone evaluations to provide instructors and students with transparent, actionable feedback throughout the delivery lifecycle.
TECHNICAL TOOLKIT
Core technologies. Click any chip to filter repositories.
Select any technology below to filter and highlight matching repositories in All Projects.
01
Agentic AI & Machine Learning
02
Backend & Full-Stack Systems
03
Cloud, Security & DevOps
LET'S CONNECT
Building secure cloud platforms and reliable AI tools.
Always happy to chat with engineering leaders, hiring teams, and builders about cloud architecture, backend systems, or applied AI.