ML/AI Engineer with 2+ years building production LLM systems, deep learning models, and scalable data pipelines delivering $229K+ verified business impact across robotics, marketing intelligence, and e-commerce. Proficient in Python, PyTorch, and TensorFlow with hands-on experience in GenAI, RAG, end-to-end MLOps, and distributed data engineering.

I came to ML through Electronics Engineering. Hardware teaches you that latency compounds, systems degrade under load, and the gap between a prototype and something production-ready is almost never just a code problem. That mindset is in everything I build.

Currently at Tatum Robotics: production ASR at <200ms, 3x edge inference speedup via INT8 quantization, and a text-to-ASL engine covering 3,000+ phrases.

"Always in Beta. Always Compounding."

I'm actively looking for my next role. If you're working on something hard and care about what gets shipped, let's talk.

Roles I'm targeting:

Industries: Worked across FinTech, Robotics, Marketing and AI. Always looking for new domains to dive in

Real systems, real constraints, real production.

Tatum Robotics - AI Software Engineer (Aug 2025 – Present)

Building production AI systems for robotic communication at the intersection of speech, language, and gesture.

• 🎙️  Whisper ASR pipeline containerized with CI/CD version control and automated quality validation, processing 500+ daily utterances at 95%+ accuracy and <200ms latency
• 🤟  Text-to-ASL translation engine on a C# (.NET) backend, mapping 3,000+ phrases to 26 hand configurations across diverse signing contexts via a gesture mapping engine
• ⚡  Post-training quantization (FP32 to INT8) benchmarked across GPU (CUDA) vs. CPU latency profiles, delivering 3x on-device inference speedup, 70% model compression, and <1% accuracy loss
• 📉  Redesigned the gesture-to-phrase mapping pipeline, reducing ASL interpretation latency by 40% and improving response consistency across varying input conditions

Crewasis AI - ML Engineer Intern (Jan 2025 - Jun 2025)

Built multimodal ML infrastructure for marketing intelligence at scale across social media platforms.

• 🧠  Fine-tuned BLIP-2 with LoRA adapters and deployed a multimodal RAG system over audio, video, and text, containerized with Docker, processing 5K+ daily social media assets
• 🚀  Scaled ETL pipeline throughput 60x (30 min to 30 sec) by deploying Python workers on AWS Lambda with Airflow triggers and automated data quality checks, saving $19K+ annually
• 🔍  Built a vector search system across 1.6M+ records integrating REST APIs (YouTube, Instagram, TikTok) with FAISS vector retrieval at sub-3s query latency, orchestrated with Kubernetes
• 📊  Validated a 29% cost advantage across 20+ A/B experiments using MLflow tracking, translating results into deployment decisions for senior leadership

Red Moments Pvt Ltd - Jr. Data Scientist (Jun 2022 - May 2023)

Data science and analytics across manufacturing and e-commerce operations in Mumbai.

• 📈  Built time-series forecasting models (Prophet + XGBoost) on 75K+ transactions with SQL-driven feature engineering, improving production planning by 23%
• 💰  Designed A/B testing frameworks translating business questions into structured recommendations, generating $100K annually with 16% inventory reduction
• 🏗️  Constructed ETL pipelines with dbt transformation workflows and CI/CD schema validation, lifting margins by 9% and producing $80K in revenue
• ⏱️  Built Tableau and Power BI dashboards with documented KPI definitions for cross-functional stakeholders, cutting reporting from 3 days to real-time and saving $30K annually

• 🔍  FinSight RAG - Hybrid RAG pipeline with MiniLM embeddings, dense/sparse retrieval, and semantic reranking over SEC 10-K filings. Benchmarks 7 retrieval strategies via an LLM-as-judge framework. 94% query success · 4.25/5 relevance · 42% latency cut · 40% API cost reduction
• 🎵  Speech Emotion Recognition - CNN-LSTM with MFCC, mel-spectrogram, and chroma feature extraction on 15K+ audio samples. 90.5% accuracy · 90.4 F1 across 8 classes · outperformed InceptionV3 baseline by 3% while training 25% faster
• 🦅  Bird Species Classification - 4 CNN architectures benchmarked on 89,885 images across 100 species. Deep VGG-style CNN vs. InceptionV3 transfer learning. 90.5% accuracy · 90.4 F1
• 🧬  NeuroDigest AI - Agentic LLM pipeline using LangChain reasoning chains to ingest multi-format sources and generate structured digests

I'm always open to conversations about interesting problems and the right opportunities. Reach out through any of the channels below.