M37Labs Logo

M37Labs Feed

Knowledge base

Direct from our engineering and research Slack channels. Raw insights, technical breakdowns, field notes, and real-time updates from the M37Labs team.

M

Team M37Labs

knowledge-base

Pyconf Resources:

Link
Mar 16, 2026 • 5:31 AM
M

Team M37Labs

knowledge-base

Building Custom AI Tools for Influencer Marketing | Arindam Paul posted on the topic |…

LinkedIn

Building Custom AI Tools for Influencer Marketing | Arindam Paul posted on the topic | LinkedIn

Influencer marketing for us has been a uniquely complicated problem. We're not a D2C only brand where you can track a creator link straight to a checkout page. D2C is a small fraction of our revenue. Most of our sales happen through general trade, modern trade, and ecommerce marketplaces, channels where attribution from an influencer video is practically impossible. So views matter, but not in isolation. A video with 50 lakh views where the audience is 18-year-olds browsing reels is worth very different from one with 15 lakh views where the audience is 30-35 year old homeowners actively thinking about ceiling fans or water purifiers. Watch time on a 10-minute YouTube comparison video signals something fundamentally different from a 30-second Instagram reel going viral. The quality of attention matters as much as the quantity. And no existing influencer marketing tool is built to think about it this way. They all optimize for reach or engagement or purchase rate as a flat number. Nobody is helping you answer: which content themes are actually driving relevant awareness and incremental sales? We work with lots of creators every year and the data was all there (views, engagement, cost per creator, shares, saves, content themes etc) sitting across spreadsheets. But the decisions on what we should do in the next quarter was still not driven by existing data So I built a V1 tool for it. Sat down with Claude last Saturday, described the problem in plain English and clearly mentioned what the inputs would be: video links, creator costs, all engagement metrics, content themes etc. Also explained what the scoring logic should be for every video Output: a scored ranking of every video weighted for what actually matters to us, theme-level analysis, and a clear view of where budget is being wasted. Under an hour later I had a working dashboard. Every creator video scored 0-100 on the scoring logic ( which is a trade secret and varies from category to category). Theme analysis showing which content angles move the needle. Backend is a Google Sheet our marketing team can maintain. V2 will have API integrations which will remove the need for this Took it and deployed it with Lovable. And there was a live V1 Tool. No developer. No SaaS. Of course, it will need months of improvement to get it to where I would want, but it is 100x better than writing a BRD for our tech team to build or getting a random SAAS product One of the real leverage of AI is building decision-making infrastructure that's specific to your business. Not generic dashboards that work the same way for a D2C skincare brand and a multi-channel appliance company. Tools that encode your logic, your weights, your definition of what "good" looks like. I'm a few months now into AI, not as a side interest but as a core working skill. It is crazy. Just start today. P.S. Very first draft of what it looked with dummy data | 72 comments on LinkedIn

www.linkedin.com

Preview image for Building Custom AI Tools for Influencer Marketing | Arindam Paul posted on the topic | LinkedIn
Link
Mar 16, 2026 • 4:20 AM
M

Team M37Labs

knowledge-base

India’s AI Wedding Buffet: Generous Portions, Political Economy Heartburn

Link
Feb 19, 2026 • 2:15 AM
M

Team M37Labs

knowledge-base

We are delighted to announce the official release of Qwen3.5, introducing the open-weight…

https://qwen.ai/blog?id=qwen3.5&utm_source=alphasignal&utm_campaign=2026-02-16&lid=w4DiEILk24n5KRcL

We are delighted to announce the official release of Qwen3.5, introducing the open-weight of the first model in the Qwen3.5 series, namely Qwen3.5-397B-A17B. As a native vision-language model, Qwen3.5-397B-A17B demonstrates outstanding results across a full range of benchmark evaluations, including reasoning, coding, agent capabilities, and multimodal understanding, empowering developers and enterprises to achieve significantly greater productivity. Built on an innovative hybrid architecture that fuses linear attention (via Gated Delta Networks) with a sparse mixture-of-experts, the model attains remarkable inference efficiency: although it comprises 397 billion total parameters, just 17 billion are activated per forward pass, optimizing both speed and cost without sacrificing capability. We have also expanded our language and dialect support from 119 to 201, providing broader accessibility and enhanced support to users around the world.
Link
Feb 17, 2026 • 7:29 AM
M

Team M37Labs

knowledge-base

PFA

PFA
image.png
Expand Image
Image
Feb 17, 2026 • 7:20 AM