Jagadish P

Engineer (~3 years) delivering measurable impact in ADAS perception systems. Improved traffic-sign recognition accuracy from ~50% to ~95% with fast retrain cycles (~15 minutes) using compact architectures and implemented U-Net–based lane segmentation to accelerate dataset readiness and validation.

Performed

PoC-level evaluations of YOLOv8, ENet, ResNet, ViT, SAM, and Grounding DINO to assess inference performance, resource requirements, and feasibility against internal KPIs.

ENGINEER– SENSOR PERCEPTION at RNTBCI (2024-06 – Present)

• Traffic-Sign Recognition – CV/DL: Evaluated public datasets (BDDK, GTSRB, NVIDIA), then migrated to a customer Zenseact-based dataset (~110k images, 30+ classes) with a 10-class / 28k-image PoC before full scale-up.
• Replaced a slow, low-accuracy YOLO CPU baseline (~2 days/run) with a custom pipeline using SqueezeNet (ImageNet) embeddings + trained Random Forest, achieving ~95% accuracy with ~2-hour full RF training.
• Reduced model artifact size (~5 GB → ~3 GB via FP32→FP16), exported an ONNX model for customer validation, and began Docker + Supervisely packaging for deployment.
• Intersection vs Roundabout Classification - ML: Trained Random Forest, MLP, and Gradient Boosting models to classify intersections vs roundabouts directly from ADASIS v2/v3 supplier data (HERE, Bosch, Continental), without external maps.
• Built Python tooling (pandas, scikit-learn) for EDA and statistical analysis to compute feature importance, drive "NO DATA" handling decisions, and choose appropriate encodings (e.g., one-hot) with the goal of improving model performance on supplier datasets.
• Enabled ~1.5× faster GT creation with ~30% fewer labelling errors, contributing ~0.5 FTE effort saved as part of a simplification activity; used in AEB false-positive ReSimulation and recognized with the AI Luminary Award.
• Evaluated pre-trained models - YOLO, ENet, ResNet, ViT, SAM, and Grounding DINO through PoC-level inference benchmarking to assess latency, resource usage, and KPI alignment for potential deployment in ADAS workflows.
• Analysed ADASIS v2/v3 supplier data (HERE, Bosch, Continental), benchmarked differences across suppliers, and generated scenarios for ADAS feature inputs (E-Horizon, Camera) using real-world datasets to support homologation and validation activities.

GRADUATE ENGINEERING TRAINEE at RNTBCI (2023-06 – 2024-05)

• Built a Python-based route-generation tool used by 20+ engineers to plan ADAS validation drives aligned with test-case coverage KPIs, improving ease-of-use and enabling non-coding teams to access route and map data effectively.
• Performed a comparative study of ADASIS v2 vs v3, mapping field and encoding differences and documenting validation and integration implications for supplier data pipelines.
• The route-generation tool was nominated for presentation at a division-level town-hall, reflecting its usefulness and impact, and was later adopted across teams with onboarding and quick-start guidance.

INTERN at RNTBCI (2023-01 – 2023-05)

• Developed a route-optimization tool that generated statistically driven drive routes for ADAS feature testing, helping homologation engineers achieve systematic test-case coverage.
• Supported ADAS validation activities by analysing route requirements and ensuring coverage completeness through data-driven route guidance.

BACHELOR'S DEGREE in MECHANICAL ENGINEERING – ANNA UNIVERSITY (2019 – 2023)

HIGHER SECONDARY (12TH) – Higher Secondary School (2019 – 2019)

SENIOR SECONDARY (10TH) – Senior Secondary School (2017 – 2017)