Ralah Aaqil

Machine Learning Engineer | Data Scientist | Software Engineer

I bring product judgment, end-to-end technical ownership, and the range to move between user needs, system design, ML pipelines, frontend, backend, infrastructure, and production reliability. I turn ambiguous ideas into clear plans and durable products, collaborate closely, and do the unglamorous hardening needed to ship tools people genuinely enjoy using.

Experience

Founding Engineer

Riffle Inc.

  • Own the platform architecture for a real-time collaborative browser-based music product, defining boundaries across Next.js clients, TypeScript services, PostgreSQL, WebSockets, CRDT documents, and media systems.
  • Designed the application domain and state architecture, including centralized transactional undo/redo with scoped action grouping and replay-safe synchronization across complex editing workflows.
  • Replaced hosted real-time providers with in-house Hocuspocus and Yjs collaboration infrastructure, implementing authenticated WebSocket sessions, document persistence, schema migrations, and reconnect recovery.
  • Architected sharing and authorization infrastructure with invitation and membership lifecycles, granular public and private permissions, access-aware discovery, live revocation, connection invalidation, and idempotent collaborator management.
  • Built end-to-end media publishing and delivery infrastructure spanning PostgreSQL data models, authenticated APIs, temporary and permanent object storage, browser and server-side MP3 encoding, validation, automated cleanup, and public delivery.
  • Created a typed AI action platform with tool schemas, manifests, client executors, persistent conversation state, action-specific interfaces, and policy-driven approvals.
  • Engineered the AI batch-execution control plane for dependency-aware destructive and non-destructive operations, isolated undo groups, conflict handling, stop and resume behavior, confirmations, failure envelopes, and telemetry.
  • Built real-time voice and multimodal interaction infrastructure covering audio capture, playback isolation, voice ducking, transcription, session lifecycle management, idle timeouts, live transcripts, and voice-aware approvals.
  • Built a provider-integrated creative services layer for stem separation, transcription, image and sample generation, writing assistance, and model-driven production workflows.
  • Owned platform reliability and recovery systems spanning authentication repair, migration safeguards, WebSocket race-condition handling, permission-aware errors, board-state cleanup, and corrupted CRDT graph recovery.
  • Built lifecycle communications infrastructure with behavior-based scenario evaluation, reusable templates, send ledgers, scheduled execution, admin previews, rollout controls, and one-click unsubscribe.
  • Established product measurement and growth infrastructure across PostHog analytics, feature flags, A/B experiments, dynamic search visibility, and structured product and AI-tool events.
  • Reduced platform complexity and latency by decomposing state stores, optimizing collaboration APIs, coordinating cache and storage locks, standardizing shared primitives, and eliminating redundant render paths.
  • Translate ambiguous founder, user, and product goals into scoped technical plans, align stakeholders on tradeoffs, and sequence large initiatives into reviewable releases.
  • Own technical delivery from architecture and implementation through code review, rollout controls, observability, incident response, and post-launch hardening.

Software Development Engineer Intern

Riffle Studio Pvt. Ltd.

  • Built and deployed API infrastructure for AI-powered songwriting tools spanning conversational assistance, lyric updates, autocomplete, rhyme generation, and configurable model selection.
  • Improved AI output quality through prompt iteration, context and history management, response formatting, and prompt logs.
  • Integrated the AI services with Supabase-backed context, WebSocket workflows, browser clients, deployment build systems, and production CORS configuration.
  • Diagnosed and resolved routing, WebSocket, module-interaction, and browser-audio issues.
  • Built movable components while helping transition the product to a new modular editor architecture.
  • Shipped Moises-powered lyric transcription flows for turning recorded audio into editable creative material.
  • Developed browser-based workflows for creating, organizing, and editing text, voice, audio, embedded media, and reusable assets.

Research Assistant

CMR Institute of Technology

  • Co-authored the IEEE paper "A Scheme to Enhance the Employability of the Differently-abled," published at ICMLANT 2024.
  • Formulated course recommendation as a reinforcement learning problem using PPO.
  • Modeled approximately 10,000 candidate profiles with a multi-discrete state representation and roughly 50 possible course recommendations.
  • Translated an accessibility and employment challenge into a recommendation framework for identifying relevant upskilling pathways.
  • Contributed across the research lifecycle, including experiment design, model evaluation, results analysis, and manuscript preparation.

Education

Master of Data Science

The University of British Columbia

2025 - 2026 | Vancouver, BC | Grade: A

Bachelor of Engineering, Artificial Intelligence & Machine Learning

CMR Institute of Technology

2021 - 2025 | Bangalore, India | GPA 8.2 / 10

Recent Projects

Beyond the Balance Sheet | Kai Analytics x ACCJC

May - Jun 2026

Python | pandas | scikit-learn | Quarto | Make | statistical modeling | clustering | reinforcement learning

  • Built reproducable data cleaning and consolidation pipelines across student metrics and financial data reports.
  • Prepared institutional health targets and evaluated financial score patterns across 115 institutions.
  • Implemented and analysed the results of OLS, mixed-effects, ridge, and lasso models with institution-grouped cross-validation.
  • Contributed peer clustering and interpretable decision-tree workflows for benchmarking colleges against comparable institutions.
  • Developed reinforcement-learning-oriented decision support pipelines while clearly separating observed associations from causal claims.
  • Produced reproducible partner-facing analysis with Python, Quarto, Make, documented schemas, and rendered reports.

DuckDB | Parquet | BM25 | FAISS | sentence-transformers | LangChain | Streamlit | S3

  • Built a retrieval-augmented product-search assistant over 4.88M Amazon software reviews and approximately 89K product records.
  • Designed scalable preprocessing and analytical storage with DuckDB and Parquet.
  • Combined BM25 lexical retrieval with FAISS semantic search and hybrid ranking.
  • Implemented S3 artifact loading, an interactive Streamlit interface, and swappable LLM answer generation.

PDF Malware Detection

Nov - Dec 2025

Python | pandera | scikit-learn | Docker | Make | Quarto | GitHub Actions

  • Built an end-to-end malicious-PDF classification pipeline covering data retrieval, validation, preprocessing, EDA, training, evaluation, and saved artifacts.
  • Compared baseline, logistic-regression, and support-vector classification approaches.
  • Added schema and data-quality validation with Pandera and integrity checks throughout preprocessing.
  • Containerized and automated the workflow with Docker, Make, Quarto, and GitHub Actions, publishing self-contained reports through CI.

Shiny for Python | Plotly | querychat | Playwright | Posit Connect

  • Built and deployed an interactive housing-planning dashboard with reactive geospatial filters, maps, charts, occupancy views, and accessibility breakdowns.
  • Developed responsive Plotly and Shiny views for exploring non-market housing across Vancouver.
  • Integrated a multi-provider LLM assistant for natural-language exploration of the underlying data.

Python | pandas | pytest | Hatch | CI | Codecov

  • Developed a Python package for memory-efficient CSV loading, dataframe comparison, string-value correction, and summary reporting.
  • Implemented dtype downcasting and categorical and sparse representations with data-integrity safeguards.
  • Added automated testing, documentation, packaging, coverage reporting, continuous integration, and release workflows.

Publication

R. Aaqil and S. P. Joy, "A Scheme to Enhance the Employability of the Differently-abled," 2024 IEEE International Conference on Machine Learning and Applied Network Technologies (ICMLANT), pp. 38-43, 2024. DOI 10.1109/ICMLANT63295.2024.00012

Technical Skills

Languages

TypeScript, JavaScript, Python, SQL, R, HTML/CSS, Bash

Product & systems

React, Next.js, Node.js, FastAPI, Supabase, PostgreSQL, Yjs, Hocuspocus, WebSockets, Zustand, OpenDAW, Web Audio, Tiptap, React Flow

Data & ML

pandas, scikit-learn, DuckDB, sentence-transformers, LangChain, RAG, statistical modeling, Plotly, Shiny, Streamlit, tensorflow, keras

Platform & quality

Docker, AWS, Vercel, GitHub Actions, PostHog, Quarto, Make, Vitest, pytest, CI/CD

Additional Activities & Experience

TELUS x TechNation Avalanche Risk Hackathon

Jan 2026

Trained a time-series forecasting model to predict avalanche risk and built a React based map demo showing predicted risk areas.

National Canadian Medical Datathon

Mar 2026

Normalized about 70K diabetes-readmission records and built a Bayesian modeling pipeline.

Peer Learning Facilitator

2021 - 2025

Supported over 60 students through coursework and peer-learning sessions.

Private Guitar Instructor

2024 - 2025

Taught about 10 students through personalized music curricula, using clear communication, tailored instruction, and individual progress tracking.