Intro0%
Skip to content
Resume

Applied AI Engineer

Toronto, OntarioAuthorized to work in Canada

Ruttansh Bhatelia

I build the systems around models: agent harnesses, orchestration, and evaluation-minded controls that make AI work reliable and reviewable, not just impressive demos.

Selected workGet in touch

Education

Honours BSc in Computer Science

Current role

AI Engineer, CredShields

Published at

IEEE MSWiM 2025

Selected work

Applied AI systems

Observability and control systems that make AI-assisted engineering easier to inspect, steer, and improve.

agentlog

GitHub

A local observability workspace that shows what AI coding agents did, where sessions broke down, and which working patterns recur across projects and tools.

I built read-only adapters for six coding tools, source-verified session analysis, and reviewable suggestions for improving AGENTS.md and skill files without changing them automatically.

  • Python
  • TypeScript
  • SQLite
  • MCP

Cutoff

LiveGitHub

A restaurant shift operations desk where a manager and browser agent coordinate orders, stock, labor, and handoffs using the same revisioned state.

I built 16 page-scoped WebMCP tools with deterministic previews, stale-write protection, and human approval before changes are adopted.

  • TypeScript
  • React
  • WebMCP

driftctl

GitHub

An opt-in continuity layer that keeps long-running Codex sessions aligned with the latest accepted goal across new prompts, interruptions, and context compactions.

I built source-linked intent tracking and deterministic conflict checks, so accepted goals and later instructions survive compaction while ambiguous changes return to the user for review.

  • Rust
  • Codex hooks
Additional projectsView projectsHide projects

codex2gpt

GitHub

A local prototype that lets ChatGPT plan and review coding work while Codex carries it out inside approved workspaces.

I built the delegation and safety boundaries around that handoff, including managed worktrees, explicit approvals, and audit records, without giving ChatGPT a general-purpose local shell.

  • Rust
  • MCP
  • OAuth

local-sec

GitHub

A local supply-chain guard that screens selected developer-tool installs before they run on the host.

I built version-maturity checks, advisory and artifact analysis, staged review for supported installs, and explicit approval gates for higher-risk flows.

  • Go
  • Supply-chain security
View moreShow lessSolProbe and LoanRef

SolProbe

GitHub

A local control room for detecting and investigating training instability before a run wastes time or compute.

I built the Apple Silicon telemetry path, repeatable fault simulations, alert workflows, and operations dashboard; NVIDIA and cloud collection remain prototype work.

  • Rust
  • FastAPI
  • Next.js

LoanRef

Live

A multi-tenant commercial loan referral platform for brokers, partners, and borrowers.

I built role-isolated workflows, a Kanban pipeline with server-enforced stage transitions, and security controls that protect tokens, forms, and tenant data.

  • TypeScript
  • Next.js
  • PostgreSQL
  • Security

Experience

Roles

Applied AI in security tooling, plus industry-partnered research at Sheridan.

202420252026

  1. AI Engineer

    Apr 2025 – Present

    CredShieldsRemoteCurrent

    • Own Applied AI systems for CredShields One: agentic workflows, orchestration, and review-friendly outputs for AI-assisted pentesting across web, cloud, API, and mobile.
    • Contribute AI components on SolidityScan for smart-contract analysis with grounded, reviewable findings.
    • Keep clear ownership boundaries between AI modules and the rest of the platform so agent work can be tested and shipped cleanly.
    • Improve reliability of LLM-assisted findings with grounding, validation, and human-review paths (fewer opaque one-shot answers).
    • Fine-tune and evaluate models and prompting setups for vulnerability-related tasks using precision/recall-style metrics, not accuracy alone.
    • Ship supporting ML tooling for feature extraction, anomaly signals, and suggested remediations alongside existing scanners.
    CredShields OneSolidityScan

  2. Machine Learning Researcher

    Jan 2025 – Apr 2025

    Sheridan CollegeOakville, Ontario

    Centre for Applied AIIndustry partner: TELUS

    • Built generative models (VAE and DDPM) to synthesize RSSI fingerprints for indoor positioning research.
    • Integrated synthetic data with regression localization pipelines (nearest neighbor, line-shifting baselines) on a 2D apartment setup.
    • Co-authored IEEE MSWiM 2025 on synthetic RSSI augmentation and hybrid localization (Zx-WKNN).
    Publication: DOI

  3. Full Stack Developer (Cross Platform)

    Sep 2024 – Dec 2024

    Sheridan CollegeOakville, Ontario

    Centre for Applied AIIndustry partner: McMaster University

    • Ported a research web app to a cross-platform Flutter client (iOS and Android).
    • Built REST APIs, Strapi CMS integration, and SQL schema with Git-based delivery.
    • Implemented authentication and data-handling controls aligned with the project's sensitive-health-data requirements.

  4. Machine Learning Researcher

    May 2024 – Aug 2024

    Sheridan CollegeOakville, Ontario

    Centre for Applied AIIndustry partner: Osteoporosis CanadaProject also involved Naryant

    • Built ML pipelines for imminent fracture-risk prediction: cleaning, feature extraction, training, and evaluation.
    • Designed a soft-voting ensemble (Random Forest, XGBoost, and logistic regression).
    • Evaluated with precision-recall, ROC-AUC, and calibration for false-negative-sensitive clinical use.

Skills

Stack

A focused view of the technologies I use to build and ship.

Programming Languages

  • Python
  • TypeScript
  • Rust
  • Go

AI & ML

  • PyTorch
  • TensorFlow
  • Hugging Face
  • Transformers
  • Pandas
  • NumPy

Web & Data Systems

  • FastAPI
  • Next.js
  • Node.js
  • PostgreSQL
  • SQLite

Infrastructure & Tooling

  • Docker
  • Git
  • Linux
  • AWS
  • Google Cloud
  • Cloudflare

Publications

Published research

Peer-reviewed work, with concise technical articles to follow.

Low-Error Indoor Positioning via Synthetic RSSI Augmentation and Zx–WKNN Hybrid Model

DOIDBLP

IEEE MSWiM 2025 · Conference paper

IEEE MSWiM 2025. Co-authored work on synthetic RSSI augmentation and hybrid indoor localization from the Sheridan and TELUS collaboration.

Research project

Undergraduate thesis

Smart-contract vulnerability detection

An undergraduate research project for finding and classifying smart-contract vulnerabilities from both source code and program structure.

I fine-tuned CodeBERT on more than 47,000 smart contracts, explored graph models over abstract syntax trees and control-flow graphs, and built template-based fixes for reentrancy and arithmetic-overflow cases.

  • Python
  • PyTorch
  • Transformers

Technical articles

Articles coming soon

Concise notes on applied AI systems, evaluation, and security.