Gagandeep Singh Email & Phone Number

Bengaluru, KA, IN

Bengaluru, Karnataka, India

• Built Face Recognition and Tracking using OpenCV, AWS Rekognition and GCP Vertex API. Allows for tracking of faces in the video from start to finish.
• Experimented with Facenet models and Cosine Similarity of Face Vector embeddings to cluster Face images in the video.
• Setup Face Upscaler pipeline using GFPGAN model to enhance poor quality videos recorded with webrtc software like Zoom or other conferencing softwares.
• Integrated Talknet-ASD model for Active Speaker Detection in videos.

Bengaluru, Karnataka, India

• Built Async Poll/Chat Aggregation pipelines using API Gateway + Kinesis + Lambda. Helps to absorb Burst Traffic and sends aggregated results in a moving window every few seconds, reducing load on the system without.
• HLS VoD Captions pipeline, streams and stores live Captions/Translations events/lines and process them post session end to generate Audio Time-Code synchronized.VTT files for the On-Demand HLS.
• In-House Video HLS Transcoder, built our own.MP4 to HLS transcoder, to save on transcode service costs and provide better alerting/reports on invalid/corrupted video uploads by our Organizers/Clients.
• Screenshot Service, lead the development of a service that generates Thumbnails for the Goldcast Events, captured using Puppeteer(JS) whenever Theme/Text Data of the Event is changed. Allows our Studio App to provide.
• PDF Slide Processor pipeline, generates.PNG images from the PDF pages uploaded to our platform, allowing clients to render Image files instead of downloading and processing entire PDF doc, improving the Frontend.
• Ticket Purchase Webhook processing workflow and Confirmation/Invoice Emails, which lead to On-Boarding of new Organizations/Clients looking to provide Paid/Exclusive entrance to their Events.

Bengaluru, Karnataka, India

• Added/Modified APIs in Backend (Django) API server and Mattermost (Go, Gorilla-Mux) fork to facilitate various Buisness features or requirements.
• Live Captions/Translations, improved our Event’s accessibility by extending the Recording Bot, piping the Live Session Audio to 3rd Party Transcription/Translation services and streaming the results via Websocket.
• Built ElasticSearch Backend to support our Attendee Search. Attendee records are enriched and streamed to ElasticSearch by utilizing PSQL Write-Ahead Logging (CDC).
• Built various AWS Lambda job processors, consuming from sources like SQS, EventBridge, S3, etc.
• Responsible for building and maintaining our RTMP In and Pre-Recorded Event offerings using AWSservices like Medialive, Mediastore, Cloudfront (CDN).

Remote

• MVP built using No Code platform Bubble to help Secure Seed Funding.
• Goldcast Recordings, bot built using Puppeteer(JS), FFmpeg, and Docker, running on GPU (CUDA)hardware accelerated intances on ECS, generating Live HLS stream of the Event which is used to provide LiteMode experience.
• Youtube, Facebook and Linkedin Live, extended Recording bot with the ability to beam the Live Stream outto External Sources using rtmp:// Protocol.
• HLS ->.MP4 Transcoder, async worker running on Fargate to process Event Recording streams andgenerate.MP4 files for clients.
• Main maintainer of Mattermost fork, our Chat and Websocket Service, running on ECS. Uses Redis PubSubto facilitate Horizontal Scaling of Websockets.

Remote

• Developing the scalable version of Goldcast, a turnkey platform for marketers and event organizers to conduct live virtual business events and conferences with measurable ROI.
• Earlier developed the early-stage MVP version of the platform. (in private beta)
• Setup the adaptive bitrate video streaming and conversion pipeline.
• Implemented a docker container running FFmpeg for HLS chunk file concatenation for easy downloads of mp4 VoD.

• Objective was to create a Chart Styles API, a feature similar to Chart Themes offered by Microsoft Office, where the user can apply custom formatting (style) to Charts in all of the applications, namely, Document.
• Explored various Chart Style API suggestions in the LibreOffice community with my mentor for a versatile solution.
• Architected the selected solution, resulting in users being able to call Create and Read operations on Chart Styles.
• Built a GUI for Chart Style operations inside the ’Chart’ panel that is available in all major LibreOffice applications.
• Implemented a requested addition to Autosum feature in the formula toolbar of LibreOffice Calc alongside the API.

• Objective was to finish the reverse-engineered Cineform decoder so that FFmpeg could play all Cineform video files.
• Acquired video sample files from multiple sources that were known not to be played correctly by FFmpeg’s decoder.
• Deduced the missing features in FFmpeg’s decoder, inclusion of which would support the playback of acquired files.
• Ported features, namely, deinterlacing and inverse-temporal transform based on the open-sourced Cineform codec.
• Multi-threaded the decoding algorithm of video frames using FFmpeg’s in-house wrapper library on POSIX threads.