a
adithyashank997

adithyashank997

India
Engels, Hindi, Tamil, Telugu
Sommige informatie wordt in het Engels weergegeven.
Over mij
A talented professional with 2 years of experience in Data solutions for industrial problems at scale. I have worked on multiple problems viz-a-viz deep learning based and traditional prediction models to reduce manual efforts while increasing the efficiency at the same time. I am looking for more such opportunities where I can create an impact on business through my work. I am proficient in developing Computer Vision based models for mainstream tasks and also experienced in creating a complete ML based pipeline for further downstream tasks. I can also build and share models for NLP tasks.... Lees meer

Skills

a
adithyashank997
adithyashank997
offline • 

Portfolio

Werkervaring

Senior ML Engineer

VeGrow • Fulltime

Apr 2024 - Present • 2 yrs 2 mos

Orchard Assessment: Built an Android app for near real-time orchard grademix assessment ■ Assessed 40 acres/day at ₹600/acre, achieving 95% accuracy and strong farmer validation ■ Generated up to ₹4 lakh additional revenue per harvest through quality-driven insights ■ Commercialized at a 4900% markup, showcasing strong unit economics and traction

AI Engineer 2

Avataar • Fulltime

Jun 2022 - Jun 2024 • 2 yrs

Pose Optimization (NeRFs): Deriving precise camera positions for training 3D NeRF models ■ Integrated multiple Pose Optimization techniques lowering the failure rate by 10%; ■ Cut the runtime by ~30%, yielding a ~30% cost reduction per unit; Link to detailed study; Diffusion Models: Efficient deployment of multiple generative models on Triton Server ~30% reduction in inference cost and time achieved by conversion to TensorRT based inf. ■ Further reduced the inference time by 2x by using distilled models; Text to Experience: LLM-driven pipeline to generate dynamic product page experiences ■ Fine tuned LLM (Mistral) model for function selection and structured output generation; ■ Functionality analogous to MCP; Generated synthetic datasets using few shot prompting;

AI Engineer

Delhivery • Fulltime

Mar 2020 - Jun 2022 • 2 yrs 3 mos

Locality Polygons: Built locality polygons for localities present for 3 states in India. ■ Increased precision of existing polygons from 70% to 80% across all the 3 states. ■ Increased the recall by 3x times in localities residing in rural areas. Utilisation Estimation: Pipeline (Video-3D) models truck utilisation and damage prediction ■ Deployed an end to end video analytics (CCTV) pipeline on 40 docks for 3 months (PoC) ■ Reduced error from existing product’s 29% to 10%; Solution’s Cost/Dock/Day - Rs50 Sorter Images: Built a classifier with to sort the packages into two categories ■ Classifier sorts packages as a Flyer/Box with 97% accuracy; impacts bill of the package ■ Catches the wrongly tagged packages and corrects the billed amount to drive revenue Warehouse Intelligence: Built a Deepstream based pipeline to optimize required workforce ■ Aggregates total idle time and flags specific compliance events across 100 cameras ■ Complete pipeline runs on a single EC2 Machine at ~ $600/ month AER: Built a model to extract different entities (ex- locality, landmark, etc) from an address. ■ Trained RoBERTa using semi supervision using distant labels, with just 3000 true samples. Private Info Masking: Built and deployed a model to mask private info from consignments. ■ For security and compliance purposes as our security team sometimes view the packages