Saurav Singh Chandel

Software developer focused on AI engineering

Intermediate Software Developer at Studix, based in St. John's, Newfoundland, Canada

I build applied AI features end to end: speech transcription, LLM analysis pipelines, document understanding with OCR and vision models, and running open-source models on our own hardware. A Joint Honours degree in Computer Science and Applied Mathematics means I'm as comfortable with the maths behind a model as with the systems that serve it.

Experience

Industry

Intermediate Software Developer

Studix (via Deel), remote for a Norway-based software company, April 2026 to present

CurrentIndustry

Recently delivered

An AI sales-call analytics pipeline: Vonage audio ingestion, faster-whisper speech recognition and LLM classification that extracts sales outcomes, customer intent and meeting commitments. Transcripts are discarded after analysis and audio is kept only for a limited period.

Implemented structured LLM prompting and inference workflows on Fireworks AI (gpt-oss-120b).
Architected a multimodal pipeline combining speech transcription, OCR, vision-language models and specialised LLMs to turn mathematics lecture videos into structured LaTeX documents and PDFs.
Built video preprocessing, chunking and context-management workflows to keep long lecture recordings efficient to process and coherent end to end.
Secured containerised Linux environments with Docker, SSH key authentication and TOTP-based MFA, and built reproducible Ubuntu environments with Packer and Ansible.
Improved reliability and fixed security and integration issues across C++ browser-streaming services, Python/Django APIs and JavaScript frontends, including Google Meet integration.

Ex co-founder

Fibly, an AI productivity workspace

2026Startup

Built OpenAI-powered features and handled infrastructure, security hardening, database reliability and payments through launch in May 2026. React, TypeScript, Supabase, Stripe, Vercel.

University

Memorial University of Newfoundland

Three roles alongside my degree

Teaching Assistant, Mathematics

Led mathematics tutorials and evaluated coursework for 100+ students, explaining complex technical concepts and giving structured feedback.

Sep 2023 to Dec 2024

Computer Science Help Centre Assistant

Helped students debug and understand programming coursework.

Sep to Dec 2024

Database Analyst

Built Selenium automation to detect and resolve email delivery failures, cutting about 30 hours of manual admin work per semester.

May to Aug 2022

Skills

AI engineering: LLM integration, speech transcription, OCR and vision models, classification, context-window management
Local and private AI: Open-source model evaluation, local inference, quantization, hardware sizing
Machine learning: scikit-learn, Stable-Baselines3, Gymnasium, NumPy, Pandas
Languages: Python, C++, TypeScript, JavaScript, Java, SQL
Systems: Django, Node.js, React, PostgreSQL, Redis, Docker, Linux, CI/CD
Mathematics: Numerical and spectral methods, SymPy, scientific computing

AI and ML projects

Space Invaders reinforcement learning

PPO agent trained on raw Atari frames with a CNN policy and frame stacking.

Stable-Baselines3, Gymnasium

Predictive modelling on imbalanced data

Compared random forest, k-nearest neighbours and SVM classifiers on 3,053 rows and 72 variables, with only 278 positive examples.

Python, scikit-learn

MedQueue

ER queue updates and wait estimates for lower-acuity patients, backed by a simulated ML model. Built with Tejaswini Gude at GenesisBuild 48.

MERN, Vercel

AI gameplay coach (concept)

Early work on a tool that analyses Valorant clips and gives coaching notes, starting with crosshair placement.

Computer vision, in progress

Education

B.Sc. Joint Honours, Computer Science and Applied Mathematics

Memorial University of Newfoundland. GPA 3.76/4.00, CS GPA 3.9. Dean's List 2022–23.

Honours thesis on black hole stability through marginally outer trapped surfaces, using symbolic and numerical spectral methods in Python and SymPy. Eigenvalues matched known solutions to relative errors as low as 10⁻⁵. Read the thesis (PDF)

2021 to 2025