Open to opportunities

sakshi goyal

@sakshigoyal

Message

Senior AI engineer building generative AI, agentic RAG, and on-device MLOps for measurable impact.

India

Message

What I'm looking for

I’m looking to build production-grade generative AI and agentic RAG systems with strong MLOps. I want ownership from architecture to monitoring, measurable outcomes, and teams that value latency, scalability, and practical experimentation.

I’m a Senior AI Engineer with 5+ years of experience across generative AI, agentic systems, NLP, computer vision, and MLOps—delivering production-ready solutions from architecture and data engineering through training, serving, and monitoring. I’ve built and shipped on-device SLMs, LangGraph-based agentic RAG pipelines, diffusion model fine-tuning, sensor fusion systems, and scalable ETL/CI/CD on AWS, with a consistent focus on latency, scalability, and business impact.

In my current role, I replaced Whisper API-based intent inference with a fine-tuned, quantized SLM deployed on-device, cutting latency ~3x and improving intent accuracy beyond 98% while removing external API costs. I also lead end-to-end AI R&D—setting technical roadmaps and latency SLAs, running safe shadow deployments, benchmarking OpenAI/Sarvam APIs against custom models, and mentoring engineers across model development and evaluation workstreams. My consulting and earlier roles add depth in Stable Diffusion LoRA, real-time TTS pipelines, dual-corpus RAG support agents, and robust MLOps fundamentals (CI/CD, Docker, Airflow, MLflow, TorchServe) that keep systems reliable in production.

Experience

Work history, roles, and key accomplishments

Current

AI/ML Engineer

Current

Lenskart

Feb 2026 - Present (4 months)

Replaced Whisper API intent inference with a fine-tuned, quantized on-device SLM, cutting latency from 600–700ms to 200–300ms (~3x) and improving intent classification accuracy to 98%+ while removing external API costs and round-trip dependencies. Built the AWS EC2 + MLflow training/serving pipeline with shadow deployments and led voice-intelligence R&D through roadmap, latency SLAs, and API vs on

On Device Inference AWS EC2 MLflow Experiment Tracking Shadow Deployments PyTorch

Technical Consultant

Dialog Matrix

Aug 2025 - Jan 2026 (5 months)

Fine-tuned Stable Diffusion with LoRA for few-shot visual concept learning, reducing required training samples from thousands to under 20 while preserving generation quality. Built an agentic Pipecat TTS pipeline with FastAPI microservices and a dual-corpus LangGraph RAG support system (ChromaDB) achieving a RAGAS score of 0.85 and ~40% ticket deflection rate.

Ragas Evaluation Stable Diffusion Pipecat Real Time TTS Streaming fastAPI LangGraph ChromaDB Agentic RAG

Data Scientist

1&1 Telecommunication SE

Sep 2025 - Oct 2025 (1 month)

Scraped and analyzed Kubernetes pod performance metrics and integrated KPIs into Grafana dashboards for real-time visibility into model performance, data drift, and infrastructure health. Enhanced fraud-login detection ML pipelines using XGBoost, improving precision and recall on production traffic.

Grafana Model Data Drift Detection Data Pipelines Fraud Detection XGBoost SQL Python

Software Developer - AI MLOps

Bertrandt GmbH

Jul 2022 - Aug 2025 (3 years 1 month)

Optimized 5+ AWS ETL workflows using Glue, PySpark, and SQL to produce analytics-ready datasets at scale. Automated CI/CD with GitHub Actions and Jenkins (80% build/test coverage) and improved development efficiency by 25%, implementing MLOps with Docker testing, Airflow training pipelines, TorchServe serving, and systematic model validation.

AWS Glue Pyspark SQL GitHub Actions Jenkins Airflow Docker MLFlow TorchServe MLOps

Software Developer - DL/NLP

STTech GmbH

Dec 2020 - Jun 2022 (1 year 6 months)

Researched and implemented CNN and Transformer architectures for out-of-distribution detection in autonomous-driving datasets using PyTorch, contributing to a peer-reviewed publication on perception reliability. Built a BERT/RoBERTa NER pipeline improving entity extraction accuracy by 28%, and implemented 2D sensor fusion (Camera/LiDAR/Radar) using a Python Kalman filter with path planning in CARL

PyTorch CNN Transformers Out Of Distribution Detection BERT RoBERTa NER Kalman Filter Sensor Fusion CARLA Sensor Fusion

Computer Vision Intern

AGT Group (R&D) GmbH

Apr 2019 - Sep 2019 (5 months)

Accelerated pose tracking by 40% using GPU-optimized optical flow and automated training/evaluation pipelines, reducing iteration cycles by 35%. Built stereo and IP camera calibration setups using OpenCV.

OpenCV Optical Flow Stereo Calibration IP Camera Calibration Data Pipelines Model Training Evaluation Automation