Headline

Senior Data Scientist @ Mercedes-Benz R&D | GenAI, VLMs & RAG | Production ML & Real-Time Systems | MLOps on GCP

About

Senior Data Scientist and ML/AI Engineer with 8+ years building production-grade AI systems — from GenAI pipelines, VLM fine-tuning, and RAG-powered applications to real-time ML infrastructure processing 150,000+ concurrent vehicle data streams. At Mercedes-Benz R&D, I work on autonomous driving effectiveness — building vision scene-mining platforms (BLIP-2, SigLIP-2, Qwen2.5-VL, CLIP), fine-tuning VLMs with LoRA, and deploying RAG + NL-to-SQL tooling that turns complex sensor data into decisions. Earlier, at Volvo Eicher, I engineered Kafka-based streaming pipelines delivering severe-fault alerts in under 5 minutes across a 150K+ vehicle fleet. I own end-to-end model lifecycles — MLflow, Kubernetes on GCP, Grafana-based observability and drift monitoring — and I enjoy leading and mentoring cross-functional ML teams. My background bridges deep mechanical-engineering domain knowledge (2 granted patents) with modern AI, which lets me translate ambiguous business problems into measurable, scalable outcomes. You can contact me +91-9993696959 or via email at devanjanmishra@gmail.com

Location

Greater Bengaluru Area

AI stack

claude code

hugging face

ollama

Devin

Milvus

vLLM

Work experience

Senior Data Scientist at Mercedes-Benz Research and Development India, Nov 2023 - Present · 2 yrs 10 mos, Bengaluru, Karnataka, India, Building GenAI and computer-vision systems for autonomous driving effectiveness. • Built a PyTorch scene-mining platform (BLIP-2, SigLIP-2, Qwen2.5-VL, CLIP) that improved scene-retrieval speed 5x over the prior manual pipeline across millions of annotated frames; indexed image vectors in pgvector (PostgreSQL). • Deployed a LoRA-fine-tuned Qwen2.5-VL image-validation service to 300+ engineers on GCP Kubernetes, replacing manual image-review workflows. • Built a RAG + NL-to-SQL pipeline (Claude Sonnet) with Redis caching, a secondary validator model, and ground-truth monitoring for bias/drift — cutting dashboard query turnaround from hours to seconds. • Generated synthetic edge-case scenes (low-light, adverse weather, degraded sensors) via a conditional generative model, integrated into an MLflow-tracked evaluation framework. • Built FastAPI services pushing live eval results into an A/B tracking store; constructed knowledge graphs over internal code repos for semantic search. Containerised via Docker, orchestrated via Flyte. • Work spans sensor fusion (LiDAR, Radar, Camera, IMU) and BEV validation for L2/L2++/L3 systems.

Deputy Manager at Volvo Eicher Commerical Vehical Limited, Jan 2022 - Oct 2023 · 1 y 9 mo, Indore, Madhya Pradesh, India, Led real-time vehicle analytics and predictive maintenance for a 150,000+ commercial-vehicle fleet. • Engineered Kafka-based streaming pipelines processing 200+ sensor parameters/min across 150,000+ vehicles, delivering severe-fault alerts in under 5 minutes; led a team of 3 engineers through architecture and deployment (FastAPI, Nginx). • Built Grafana observability tracking model performance, statistical data drift, and anomaly-model degradation across the live fleet. • Forecasted warehouse demand with time-series ML (Scikit-learn, TensorFlow) to reduce critical spare-parts stockouts; built live sensor-stream monitoring for early-warning engine-health alerts. • Designed geo-map and heatmap dashboards (Folium, Apache Superset); mentored 2 analysts.

Lead Engineer at GENERAL MOTORS TECHNICAL CENTRE INDIA PRIVATE LIMITED, Apr 2021 - Jan 2022 · 10 mos, Bangalore Urban, Karnataka, India, Applied ML to vehicle safety and structural design optimisation. • Cut model-validation cycles 50% with a stacked ensemble (Random Forest, SVM, ANN, XGBoost) predicting safety injury criteria across design variants; deployed as a self-serve internal tool. • Built Python regression models predicting structural performance curves from simulation data, eliminating multiple physical prototype iterations. • Led multi-disciplinary optimisation using surrogate modelling and Design of Experiments; authored ML methodology guidelines adopted as team-wide standards; trained/mentored engineers across CAE teams.

System Engineer at GENERAL MOTORS TECHNICAL CENTRE INDIA PRIVATE LIMITED, Oct 2019 - Mar 2021 · 1 yr 6 mos, Bangalore Urban, Karnataka, India, ● Trained 2 other CAE teams and a total 7 people, on Machine Learning methods for CAE based design and optimization. ● Authored three global method guidelines (plus one under review), two internal research reports on areas of countermeasure efficiency, placement strategies and RRAB deployment.

Senior Engineer at GENERAL MOTORS TECHNICAL CENTRE INDIA PRIVATE LIMITED, Apr 2018 - Sep 2019 · 1 y 5 mo, Bengaluru, Karnataka, India, ● 33% reduction in hardware testing over the course of 4 years, through correlation- prediction improvement, acceleration -time regression curves and lead design variation studies for load case. ● Developing Inhouse Procedures for quicker overall execution for balancing and optimizing static and multiple impact loadcases into a single routine using Design of Experiments and Kriging

Senior Lead at General Motors Technical Centre India (via TCS), Jul 2016 - May 2018 · 1 y 10 mo, Bengaluru, Built CAE crash simulation models to predict crash performance along with data, enabling early-stage design validation and reducing design iterations for Crash requirements.

Education

Post Graduate Program, Artificial Intelligence & Machine Learning, Texas McCombs School of Business

Bachelor of Technology - BTech, Mechanical Engineering, Maulana Azad National Institute of Technology

Public links

github: https://github.com/devanjanmishra