All Hands Meeting (Group 1: R&D)(Recorded using Fireflies)

🕒 Date: Tue Oct 31 at 2:30 PM UTC

⌛ Duration: 86.00 mins

🗒 Transcript: See full transcript

🔊 Audio: Listen to audio

Summary

Kapil Agrawal greets the group, followed by Akhil Bhiwal and Heril. They discuss Halloween celebrations and running out of chocolates. The conversation shifts to music production, with Ashwin N sharing his interest and collaboration opportunities. Shahab introduces himself as a new member who loves football and traveling. Harold joins the sales team and shares his passion for food. The group discusses bars, discounts, and video meetings before ending the conversation.
During this part of the transcript, Akhil Bhiwal expresses gratitude to the measurement team for their excellent presentation. Michael Lee highlights a promising call with Grin and discusses the potential benefits of securing them as a client. The ticket size for Drain is also discussed. Several team members give shout-outs and thank yous to colleagues for their contributions, including Rajesh, Akash, and Nitin. Phalgun Guduthur provides updates on recent releases and achievements, such as surpassing 300,000 connected accounts.
Phalgun Guduthur provides updates on the number of accounts, API hits, and product usage. Anujith asks about key product metrics and the importance of tracking Connected Accounts. Phalgun explains that Active Connected Accounts is more significant and mentions issues with Bintango. The team discusses measurement trends and outliers in different metrics, including profile analytics and content API hits. They move on to design updates, such as adding contact options in search and sorting features for marketers. They also mention upcoming projects like onboarding tools, handling Instagram stories, usability updates, promo codes, and fingerprinting design. Akhil Bhiwal shares visual design progress related to engagement rate QA, Thousand Faces website launch, social media tasks with Swastik and Drew, newsletter template development with Sai, open graphs implementation, among other minor tasks. Mohit reports new onboarded clients (Viral Nation) and pending approvals for chat GPT plugin from OpenAI side and Shopify app approval from Nathan's efforts. He also mentions marketing plans for the chargebd plugin. Metrics remain consistent except for an issue with webhook delivery which has been fixed by introducing alerts. R&D items related to logo extraction from images/video speech extraction are being worked upon by the engineering team under Dinesh Rajpoot's guidance who explains the delay in Shopify approval process due to staggered reviews but expects it to go live soon along with upcoming work on promo code integration that will make the product sellable.
Dinesh Rajpoot discussed customer requirements they are currently working on. There was a discussion about the approval process for accessing the Chat GBD plugin and the need for improvements in relevance when searching for specific influencers. The team planned to release a video showcasing the plugin and gather feedback from users. They also talked about designing AI-first interfaces and how it will change product design principles. There were discussions about declining metrics, API calls, and active connected accounts. The team considered fine-tuning results based on expected queries and prompt guides. They also discussed how logo extraction is done without storing all content or crawling it like YouTube does with search results.
Akhil Bhiwal asks about indexing and crawling social media sites, particularly for brand mentions or logo mentions. Arvind explains that they are in the early stages of evaluating technical feasibilities and solution approaches for single content items, with plans to expand to larger volumes later. Akhil also asks about how social listening API works on text, and Arvind explains that filter information is passed to IQD (Intelligent Query Dispatcher) which performs the activity and provides a response.
Kapil Agrawal discusses the process of extracting image or video details from influencer or brand profiles. They currently analyze past content and extract relevant metrics, but it's based on individual content pieces. Akhil clarifies that they get a list of URLs or influencer profiles and process each content piece individually using vendor APIs.
Mohit raises concerns about the challenge of providing insights for influencers due to the need to crawl profile images and perform other tasks on-the-fly. Michael Lee asks about the source of their AI models, whether they use open-source models like speech-to-text or machine vision models for tasks like logo detection. Kapil explains that they primarily use open-source models hosted on platforms like Hugging Face or Google.
Michael Lee further explores whether there will be costs associated with AI model usage, and Kapil confirms that most costs come from compute resources rather than AI model acquisition. He mentions that OpenAI APIs are used for features like relevance and purchase intent analysis.
The discussion briefly touches upon other companies like Entropic (an emotion AI company), but no detailed comparison between OpenAI and others is provided.
The conversation shifts towards broader topics such as form factors in AI interfaces, mass adoption of open-source AI models, competitive landscape analysis in a world where AI models are widely distributed for free, etc.
Masroor P Mohamed updates on ongoing activities related to cloud subscription audits aimed at ensuring proper tool usage without idle accounts. He also mentions the domain email warm-up process to improve email deliverability and open rates. The team has faced challenges with emails landing in promotions or spam tabs, but they are using a tool for warming up emails and improving their reputation.
Phalgun Guduthur discusses key performance indicators (KPIs) for the dashboard products. During the initial stages of search and campaign tracking, KPIs focused on the number of customers who successfully analyzed creator profiles and contacted them.
Note: This summary only covers part 5 of the transcript.