Skip to content
View rishabh2301's full-sized avatar

Block or report rishabh2301

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
rishabh2301/README.md

Hi πŸ‘‹, I'm Rishabh Saraf

Senior AI Engineer @ TII


πŸ‘¨β€πŸ’» About Me

  • πŸ’Ό Senior AI Engineer at Technology Innovation Institute (TII), Abu Dhabi, working on the Falcon LLM and Falcon Audio teams
  • 🧠 6+ years across LLM pretraining, audio-language models, search & ranking, and NLP
  • πŸ“„ Publications at ASRU 2025 and IEEE TPAMI, plus 2 US patents
  • πŸŽ“ IIT Dhanbad β€” Integrated M.Tech, Mathematics & Computing (9.19/10, Rank 3)

πŸ”­ What I Work On

  • πŸ—οΈ LLM pretraining β€” architecture design, GQA / SwiGLU, Triton kernels, HF checkpoint conversion
  • πŸ”Š Audio-language models β€” speech, music & sound understanding (MMAU, AirBench)
  • πŸš€ Inference optimization β€” throughput & latency for long-form audio
  • πŸ”Ž Search, ranking & retrieval β€” intent models, pairwise ranking, embedding retrieval

πŸ“Š Professional Journey

🏒 Company πŸ’Ό Role ⏱️ Duration ✨ Highlights
TII Senior AI Engineer 2024 – Present Falcon3 pretraining β€’ Triton kernels β€’ Falcon3-Audio (+10 pts on MMAU/AirBench)
Swiggy Data Scientist II β€” Search 2022 – 2024 Query intent model β€’ Pairwise ranking β€’ Siamese retrieval β€’ πŸ† MVP Award
Rakuten Institute of Technology Data Scientist II β€” R&D 2020 – 2022 NLU FAQ matching β€’ BERT/RoBERTa embeddings at 165M scale
IBM India Software Labs Data Science Intern β€” Data & AI Jan – Jul 2020 Pair-analysis micro-service for InfoSphere MDM β€’ Dockerised REST APIs β€’ 2 US patents
National University of Singapore Research Intern May – Jul 2019 Dense video captioning & segmentation benchmarks β€’ IEEE TPAMI 2022
University of Manitoba (Mitacs Globalink) Research Intern May – Jul 2018 Multilevel multiple imputation on longitudinal healthcare data

πŸ“„ Publications & Patents

  • Competitive Audio-Language Models with Data-Efficient Single-Stage Training on Public Data β€” ASRU 2025
  • Transferring Knowledge from Text to Video: Zero-Shot Anticipation for Procedural Actions β€” IEEE TPAMI 2022
  • Automatic Tuning of Thresholds and Weights for Pair Analysis in an MDM System β€” US Patent 11681671B2
  • Pair Selection for Entity Resolution Analysis β€” US Patent App. 20220035777A1

πŸ’» Tech Arsenal

πŸ€– LLMs & Training

Falcon3 β€’ Whisper β€’ GQA β€’ SwiGLU β€’ Tensor Parallelism β€’ LoRA/PEFT β€’ TRL

πŸ“š Libraries & Frameworks

PyTorch β€’ Triton β€’ Transformers β€’ Accelerate β€’ Datasets β€’ TensorFlow β€’ PySpark β€’ Scikit-Learn

βš™οΈ Serving & MLOps

TensorFlow Serving β€’ Seldon β€’ FastAPI β€’ Docker β€’ Grafana

☁️ Platforms & Data

AWS β€’ GCP β€’ Databricks β€’ Redis β€’ Elasticsearch β€’ Hive β€’ Snowflake

πŸ† Honors

  • πŸ₯‡ Most Valuable Player Award β€” Swiggy (2023)
  • πŸ‡¨πŸ‡¦ Mitacs Globalink Research Intern β€” top 800 of 10,000 applicants (2018)
  • πŸŽ“ Academic Excellence Award, IIT Dhanbad β€” top 2 for 3 consecutive years
  • πŸ“ JEE Advanced AIR 5035 Β· JEE Main AIR 6585 (2015)

Pinned Loading

  1. triton-llm-kernels triton-llm-kernels Public

    Tutorial: RMSNorm and SwiGLU kernels in Triton, forward and backward.

    Python

  2. multilevel-imputation-school-wellbeing multilevel-imputation-school-wellbeing Public

    Mitacs Globalink internship (2018): multilevel multiple imputation and linear mixed-effects modelling of a longitudinal school wellbeing trial, in R

    R

  3. tasty-procedural-video tasty-procedural-video Public

    Research code from my NUS internship on the Tasty Videos dataset: ResNet-50/I3D features, S2VT captioning, and weakly supervised step segmentation (IEEE TPAMI 2022)

    Python