Get in touch

Ritin Raveendran

ritin.raveend@gmail.com · LinkedIn · GitHub

Summary

  • Research engineer working at the intersection of speech and audio machine learning, on-device AI for consumer electronics, and LLM-based multi-agent systems for large-scale software engineering.
  • Author of BLEEP, a word-level broadcast profanity speech dataset (accepted at IEEE SLT 2026, Demo Track), covering vocabulary design, crowdsourced speaker elicitation, automated quality control, forced alignment, and ASR benchmarking.
  • Designed and built a multi-agent system that migrated ~4,000 downstream GStreamer patches across versions, with only ~46 needing manual conflict resolution, cutting migration effort from ~10 to ~1 person-month.
  • Lead a 10-member cross-team task force driving the adoption of AI and agentic development tools across a 500+ engineer department.
  • Background in timing-sensitive hardware–software measurement, reproducible dataset pipelines, and academic–industry research collaboration.

Publications

  • R. R. Kasthuri, “BLEEP: A Broadcast Audio Dataset for Word-Level Profanity Detection,” IEEE Spoken Language Technology Workshop (SLT), Demo Track, Palermo, Italy, Dec 2026 (accepted, to be presented). Dataset & code: github.com/ritin-kasthuri/bleep-dataset

Education

Georgia Institute of Technology, Atlanta, USA

Master of Science, Electrical & Computer Engineering · Aug 2018 – Dec 2019

Vellore Institute of Technology, Vellore, India

Bachelor of Technology, Electronics & Computer Engineering · Aug 2014 – Jul 2018

Independent Research

BLEEP: Broadcast Audio Dataset for Word-Level Profanity Detection

2026

  • Built an 8,312-clip word-level English speech dataset from 86 US and UK speakers, covering 12 base profanity terms, 8 spoken variants, and 29 phonetically confusable clean “hard negatives” in neutral and expressive registers.
  • Designed the vocabulary around broadcast regulation (FCC and Ofcom severity tiers), selecting hard negatives by phoneme distance in CMUdict and filtering by SUBTLEX-US/UK frequency to retain words common in broadcast speech.
  • Built a web-based elicitation interface and recruited 95 speakers via Prolific, with consent, microphone check, demographic metadata (accent, region, age, gender), and randomised neutral and expressive recording blocks.
  • Implemented automated quality control using Silero VAD for speech localisation, SNR estimation against per-speaker silence recordings, and Whisper large-v3 content verification, followed by manual review (9 speakers and 109 clips excluded).
  • Produced a fixed-onset version of the dataset with the Montreal Forced Aligner and a custom profanity pronunciation dictionary, reducing variation from reaction time and recording latency.
  • Benchmarked wav2vec2-base/large, HuBERT-large, and Whisper-tiny/base; introduced a ΔCER profanity-penalty metric showing clean-speech SSL models err 0.11–0.13 CER more on profanity than on matched clean words, while Whisper's gap is near zero.
  • Obtained consent for voice as biometric data with disclosure of re-identification risk; released the dataset and code under a gated, research-only data use agreement.

Experience

Research Engineer — LG Soft India

Bengaluru, India · Feb 2023 – Present

Multi-Agent Framework for GStreamer Patch Migration

Mar 2026 – Present

  • Designed and built a multi-agent orchestrator on Amazon Kiro to port LG's downstream GStreamer patch series from 1.24.10 to 1.28.4, a migration that conventionally takes ~10 person-months.
  • Decomposed the migration into per-patch pipelines of specialised agents for conflict detection, API-compatibility analysis, semantic auditing, blast-radius analysis, conflict resolution, and test verification.
  • Developed a deep patch analyser that generated deep analyses for all ~4,000 patches, stored in a per-patch artifact store keyed by commit SHA and used as grounded input to agents during rebase.
  • Added a retrieval-augmented (RAG) memory and git rerere to record prior conflict resolutions, so resolved conflicts are reused automatically in future migrations.
  • Migrated ~4,000 patches with only ~46 requiring manual conflict resolution, completing the migration in ~1 person-month.

AI Task Team Leader, AX Initiative

Apr 2026 – Present

  • Lead a 10-member task force, with one representative from each team in the department, owning the distribution of AI tools and the training of engineers to use them effectively across a 500+ engineer department.
  • Designed and delivered training sessions on AI coding assistants and agentic workflows; coordinated department-wide rollout of and access to Amazon Kiro and Claude.
  • Organised a department AI hackathon to drive hands-on adoption of agentic development practices.
  • Co-created a prompt bank to standardise effective AI-assisted workflows, and designed and ran internal surveys tracking AI-tool adoption and usage patterns to guide enablement priorities.

AI-based Live Audio Content Filtering for Consumer TVs

Oct 2025 – Mar 2026

  • Led the feasibility study for an AI-driven profanity detection system that mutes offensive words in real-time audio on LG webOS TVs, working with a cross-functional team.
  • Architected a real-time pipeline integrating streaming speech analysis with the TV media stack, including modifications to GStreamer-based playback pipelines and webOS media services.
  • Evaluated lightweight edge AI models under strict on-device constraints, focusing on low-latency inference and robustness to broadcast noise.
  • Identified that ASR models trained on clean read speech systematically under-recognise profanity; after the product track was discontinued, continued the work as independent research that became the BLEEP dataset.

Academic–Industry Collaboration: Audio–Video Quality Assessment (AVQA)

Jun 2024 – Dec 2025

  • Research collaboration with the Indian Institute of Science (IISc), Bangalore, on AI-based audio–video quality assessment for consumer TV testing pipelines.
  • Designed and implemented mechanisms to capture and route TV audio–video outputs into PC-based analysis pipelines using capture cards, enabling controlled benchmarking and dataset creation.
  • Evaluated the audio and video quality models proposed by academic collaborators and suggested architectural and training refinements, including embedding-space analysis and mono/stereo handling.
  • Designed multi-stage dataset pipelines combining real TV outputs with synthetically generated audio–video degradations, used for systematic evaluation, fine-tuning, and model refinement.

Audio–Visual Synchronisation and Timing Measurement for HbbTV Systems

Feb 2023 – Jun 2024

  • Owned the end-to-end investigation, measurement, and validation of audio–visual synchronisation and timing behaviour for HbbTV-enabled TV features in European broadcast environments.
  • Designed, assembled, and validated phototransistor-based optical sensing and analog audio setups with microcontrollers to measure display timing with millisecond-level temporal resolution.
  • Developed and executed measurement protocols for switch-time characterisation in targeted-advertisement scenarios, using QR-based visual markers and audio-decoder events aligned with HbbTV specifications.
  • Scaled the measurement harness with parallel processing and task queues, achieving a 4× throughput improvement.

Full-Stack Developer — Trustgrid.AI

Bengaluru, India · Feb 2022 – Feb 2023

  • Integrated RASA-based conversational models into production chatbot workflows, gaining hands-on exposure to applied ML.
  • Built frontend and backend features with React, Node.js/Express, and MySQL, including real-time voice/video interaction, screen recording, file handling, and session management.

Co-Founder — Vantir

Bengaluru, India · Dec 2019 – Feb 2022

  • Designed an educational framework (KASE: Knowledge, Attitude, Skill, Exposure) for early- and mid-career professionals, focused on holistic career development.
  • Conducted structured customer discovery and needs analysis to understand skill gaps and learning requirements in working professionals.

RF Firmware Intern — Intel

Bengaluru, India · Jan 2018 – Jul 2018

  • Developed Python scripts to automate RF validation and test workflows in lab environments.
  • Assisted in debugging physical-layer RF firmware issues, working closely with RF and systems engineers.
  • Analysed timing, interference, and signal integrity with oscilloscopes and Rohde & Schwarz spectrum analysers.

Teaching & Mentoring

  • Design and run the department's SWPCT software-proficiency test every two weeks for a 500+ member engineering department, including question-paper creation and evaluation criteria; earlier ran a mock-exam programme (6 exams in 2025).
  • Mentored 25+ employees through multiple structured 8-week preparation and Q&A programmes on data structures and algorithms.
  • Edforall (Undergraduate): Designed curriculum as Education Head to teach English to Tamil-speaking school children.

Professional Activities

  • Accepted, to be presented: IEEE Spoken Language Technology Workshop (SLT), Palermo, Italy (Dec 2026)
  • Technical talk: “Understanding AI using the Embedding Space” (LG, Aug 2025)
  • Technical talk: “Audio in the Age of Generative AI” (LG, Nov 2025)

Awards and Certifications

  • LG High Achiever Award (2024–25), recognised for technical excellence
  • LG Software Proficiency Certification (SWPCT / LSET), ranked in the top 1% of LG software engineers
  • HiPo (High Potential) member, LG (2025, 2026)

Technical Skills

  • Languages: Python, C/C++, JavaScript
  • ML/DL: PyTorch, TensorFlow, TorchServe, Weights & Biases, NumPy, SciPy, Pandas
  • Speech & Audio: Whisper, wav2vec2, HuBERT, Montreal Forced Aligner, Silero VAD, Librosa, CMUdict
  • Agentic AI: Multi-agent orchestration, Amazon Kiro, Claude
  • Video/Audio Processing: GStreamer, FFmpeg, OpenCV, TSDuck
  • Web, Backend & Databases: Flask, React, Node.js/Express, Redux; MySQL, MongoDB, Redis
  • MLOps / Infra: Docker, Kubernetes, Git, Ubuntu, Apache, VMs
  • Hardware & Sensors: NVIDIA GPUs (A4500, GV100, 3070 Ti), Raspberry Pi 5, Arduino Due, NodeMCU; optical/IR/audio sensing