Apr 2024 – Aug 2024 · 4 mos · Bengaluru, Karnataka, India · On-site

• Curl Hg gets acquired by Amphora, a leading ETRM player, to strengthen its AI-led automation stack. The rebuilt platform now powers document intelligence across trade workflows. I created the AI roadmap for integrating Sara to augment Amphora’s core offerings.

Jul 2022 – Apr 2024 · 1 yr 9 mos · Bengaluru, Karnataka, India · On-site

• Revamped the AI backbone of Sara CF, an information extraction platform for commodity trading’s buy/sell/ship flows. Introduced a domain-first GenAI stack with a fine-tuned LLM engine (via LoRA) and unified taxonomy.
• Led a 15-member Data Science + MLOps team, brought in a competency matrix to clarify roles and build deep learning and LLMOps skills, and helped shift a tech-hesitant org toward a data-first culture.
• Built a secure, on-prem LLMOps framework with distributed inferencing and integrated LLMs into production (including a Microsoft Teams QA bot). The result: reduced manual overhead, cleaner data, and a strong ROI story—culminating in Sara CF’s acquisition by Amphora ETRM.

Nov 2021 – Jul 2022 · 8 mos · Bengaluru, Karnataka, India

Jun 2020 – May 2021 · 11 mos

• As a DL Evangelist and DL Solution architect, I spearhead ML to DL transitioning of various models in Swiggy and have taken several CV and NLP models to production.
• Graph Convolutional Networks (GCN) for Information Extraction: Menu Digitization is a Structured Information Extraction task to extract entities from a menu. Built on top of AWS Textract OCR output, graphs are created using visual, semantic (NLP), and structural embeddings, and entity extraction is done using a GCN-based model for node classification.
• Image-to-Ingredients: Predicting dish ingredients from a food image as a Visual Linguistic task, using transformers (NLP). This involves curating a Recipe-Image dataset for Indian cuisine.
• DeepGastro is an edge-inferencing rig for food classification at Swiggy's Kitchens, armed with Jetson Nano, Raspberry Pi 4, along with RGB and Thermal cameras. I trained a Wide-Sliced ResNet model to identify dishes and reduce wrong orders, and optimized it for edge deployment.
• One-shot video generation from a single image: Using 3D Image In-painting, generated catalog videos to increase item desirability.
• Based on RetinaFace architecture, we trained and deployed a real-time mask detection model within three weeks of the lockdown for Swiggy Delivery Executives (DEs). I created a synthetic dataset to augment faces with different kinds of occlusion (mask, kerchief, hand).
• We automated Contactless delivery checks for DEs using Yolo V3- based person detection model. A gear and attire check was implemented, based on EfficentDet architecture, to enforce if DEs were using authorized Swiggy bags and t-shirts.
• Image Quality Assessment for Catalog Images: Enforcing image quality guardrails on photoshoot images using BRISQUE and NIMA model along with image segmentation and saliency detection.

Jan 2019 – Jun 2020 · 1 yr 5 mos

• My startup, Kint.io got acqui-hired by Swiggy.
• https://tech.economictimes.indiatimes.com/news/startups/swiggy-acquires-bengaluru-based-ai-startup-kint-io/67829829

Oct 2014 – Jan 2019 · 4 yrs 3 mos · Greater Bengaluru Area · On-site

• Co-founded Kint, an AI startup for sports video-to-game analytics, which was acqui-hired by Swiggy, demonstrating entrepreneurial leadership and strategic vision.
• Developed a Game Understanding Framework (GUF) with an overall accuracy of 89.3%, involving player detection, tracking, and re-identification, showcasing technical expertise and innovation.
• Built a 2.5D world of the game from video footage using depth perception, planar surface detection, and instance segmentation, enhancing game analysis accuracy and user experience

Mar 2012 – Jul 2013 · 1 yr 4 mos · Bangalore

• Worked as an Innovation consultant/facilitator in various engagements with Mahindra & Mahindra, Tanishq — Titan Jewelry, ITC, MTR Foods etc.
• Highlighting a few of the engagements:
• Collaborated with the marketing team of India's first electric car
• Worked with the leadership, zonal and area sales teams of a HCV (Heavy Commercial Vehicles) manufacturer. The purpose of the intervention was to find the key blocks and enablers for their sales strategy and to bring synergy between the leadership and sales teams.
• Facilitated ideation workshops for frontline workers and first-level managers to encourage newer ways of problem-solving in the factory floor. This has helped them to come up with lateral solutions for improving their processes.
• Helped facilitate a workshop with the leadership team of a Micro-financing company to create a shared aspiration, aligning their goals and the vision for the company.
• Anchored the secondary research for the book ‘Orbit Shifting Innovation’. Also identified and dialogued with unique innovators in various fields to share their stories and learnings in the book.(http://erehwonconsulting.com/book.php)

Apr 2011 – Sep 2011 · 5 mos · Bangalore

• Spearheaded the design of marketing strategies for a GE Healthcare ultrasound product, Venue OB, through ethnographic user research, involving Gynecologists, across Chennai, Coimbatore and Bangalore.
• Interfaced with end users using multiple research methodologies such as Phased interviews, In-depth interviews and Diagnostic Facility Mapping to identify the value perception of the product.
• Developed and presented to the executive board, go-to market strategies for product positioning based on customer profiling and targeted messaging.

Jun 2008 – Jun 2010 · 2 yrs · Hyderabad

• Engagement: Global Consumer Support – Microsoft
• Designed and developed a .NET based customer care support solution, servicing more than 10,000 customer care agents across the globe.
• Led the initiative to automate the customer care solution for Microsoft Windows and XBOX support groups.
• Interacting with end users in Manila (Philippines) and developed business requirements for an Agent Utilization Tracking system.
• Designed a business process guidance tool, for customer call agents, collaborating with cross-functional teams, which reduced their training period from 2 months to 3 weeks.
• Fine-tuned the Microsoft Customer Care framework to reduce the average customer call time by 15%.
• Provided hands-on training to customer call agents through demos and analyzed user stories to develop better UI features to improve usability.
• Built a knowledge base and gathered best practices for the implementation of the existing framework and business processes involved in customer care.
• Received the Customer Partnership Champions award in January 2010 for winning customer goodwill and trust with GCS, Microsoft
• Engagement: Venetian Casino, Las Vegas and Singapore
• Developed a single point of access application for Sands Marina to consolidate their Lodging Management, Restaurant, Concierge, Job Dispatch, Sales & Catering and Ticketing Applications.
• Created prototypes to host and automate varied applications to validate customer requirements and to prove stability of the proposed solution
• Integrated with their Customer Relationship Management (CRM) system to provide user-specific information for improving sales through up-sale and cross-sale opportunities

Jul 2007 – Jun 2008 · 11 mos · Bangalore

• Analyzed customer requirements to design and develop new UI features of Rational Portfolio Manager, an Enterprise Portfolio Management tool.
• Managed the import/export functionality of the product and redesigned the component to support future versions of Microsoft Project.
• Conducted risk analysis while handling change requests and escalations from clients regarding data migration issues.