Headline

Applied ML Engineer · Speech AI | Fine-tuning ASR/TTS · Multilingual NLP · Inference Optimization | AWS AI&ML Scholar | Ex-IIT BHU | Bengaluru

About

🚀 Machine Learning Engineer | Speech AI | NLP | Generative AI I build production-grade AI systems at the intersection of Speech, Language, and Generative AI—from fine-tuning ASR/TTS models to optimizing inference pipelines and deploying scalable ML products used by real users. At Voicing.ai, I improved multilingual speech translation quality by +19 BLEU points through selective encoder-layer freezing, custom Focal + Coverage loss functions, and domain-adaptive training. I also developed transcription reliability pipelines that significantly reduced hallucinations using n-gram blocking, repetition penalties, and word-level confidence scoring. At Sarvam AI, I reduced TTS streaming latency from several seconds to <500ms through model quantization and architecture optimization, achieving a 3–4× speedup. I also deployed production-scale Whisper ASR systems on GCP using NVIDIA Triton and Ray Serve. Core Expertise • Speech Recognition (ASR) • Text-to-Speech (TTS) • Speech Translation • NLP & LLM Fine-Tuning • RAG Systems & AI Agents • Model Optimization & Quantization • Distributed Inference & MLOps • Production ML Systems Tech Stack PyTorch • Hugging Face • Transformers • vLLM • ONNX • NVIDIA Triton • Ray Serve • FastAPI • LangChain • AWS SageMaker • Docker • GCP I focus on building AI systems that perform not only in research benchmarks but also in production, evaluating models across WER, CER, BLEU, MOS, throughput, and P50/P75/P99 latency. 🏆 AWS AI & ML Scholar 2023 — Ranked among the Top 10 of 1,000 scholars across 80 countries. 📍 Open to opportunities as a Machine Learning Engineer, Applied Scientist, NLP Engineer, Speech AI Engineer, Research Engineer, or AI Engineer in Bengaluru (Hybrid/On-site).

Location

Bengaluru

AI stack

anthropic apis

antigravity

cursor

elevenlabs

grok

hugging face

Work experience

Machine Learning Engineer at Voicing AI, Sep 2025 - Present · 1 yr, Bengaluru, •Fine-tuned multilingual Speech-to-Text translation models using selective encoder-layer freezing and custom Focal + Coverage loss, improving Synthetic BLEU by +19.46 pts (40.12 → 59.58) and FLEURS BLEU by +13.32 pts (23.67 → 36.99). •Designed a transcription reliability pipeline combining probability-based filtering, teacher-forced second-pass validation, and word-level confidence scoring — reducing hallucinations via n-gram blocking, repetition penalties, and beam-search optimization. •Benchmarked 10+ open-source ASR/TTS models at production scale on high-end GPUs across WER, BLEU, CER, MOS, throughput, and P50/P75/P99 latency to drive model selection and deployment decisions. •Improved TTS quality and accent neutralization through domain-adaptive training on targeted speech corpora.

AI Developer at SarvM.AI, Oct 2024 - Aug 2025 · 11 mos, Bengaluru, • Optimized a real-time TTS inference pipeline using model quantization and architecture tuning, achieving a 3–4× inference speedup and reducing WebSocket audio streaming latency from several seconds to 400–500ms. Fine-tuned Whisper Large v3 on a custom domain-specific dataset using QLoRA (Quantized Low-Rank Adaptation) + SFT, improving transcription accuracy while reducing GPU memory footprint for production deployment. • Deployed production Whisper on GCP using NVIDIA Triton Inference Server + Ray Serve, maximizing GPU utilization and enabling high-concurrency real-time transcription. • Integrated speech recognition pipeline with Android and iOS apps, achieving a 20× efficiency improvement vs. prior LLM-based automation.

AI/ML Engineering Intern at SarvM.AI, Jun 2024 - Oct 2024 · 5 mos, • Generated synthetic audio data for multiple Indian language accents to expand speech recognition training datasets. • Supported model deployment pipelines ensuring smooth integration and operational stability.

Data Science Intern at iNeuron.ai, Mar 2023 - Sep 2023 · 7 mos, Bengaluru, • Built and deployed CorrSpell — an end-to-end NLP spelling correction tool — from design docs through production deployment. • Authored architecture and design documentation; created user-facing wireframes following industry standards.

AWS AI & ML Scholarship Recipient at Amazon Web Services (AWS), Oct 2022 - Oct 2023 · 1 yr 1 mo, Remote, • Successfully led and executed various machine learning projects, delivering substantial accuracy improvements in bike sharing demand prediction, handwritten digits classification, landmark Classification & tagging for social media, and developed a comprehensive ML workflow for Scones Unlimited on AWS SageMaker. • I Was among the top 10 finish-line winners in the first cohort of 1000 AWS AI and ML scholars from 80 countries , and for that, I was chosen to get a reward.

Education

B-Tech, Computer Science and Engineering, Rajiv Gandhi Institute of Petroleum Technology (RGIPT)

Ceramic Sciences and Engineering, Indian Institute of Technology (Banaras Hindu University), Varanasi