Mohit Kale

Building AI products. Exploring startup ecosystem. Open source when I can. 17+ yrs in tech. Sharing what works, breaks and gets overhyped.

[Mohit Kale @mohitkale_]

I have spent 17+ years building software.

These days I am mostly interested in one question:
What can one person build now that required an entire team a few years ago?

I build AI products, open-source tools and experiments around that question.

Some work. Some fail. I share both.

Currently building/exploring:
• Reel Studio + Voiceforge
• AI + developer tooling
• QuantX Algo
• Startup experiments
• TRPCrew with my Troo mascot

GitHub ↓

[Mohit Kale @mohitkale_]

Github profile: https://t.co/tCQMOl6KSi

[Mohit Kale @mohitkale_]

I have spent 17+ years building software.

These days I am mostly interested in one question:
What can one person build now that required an entire team a few years ago?

I build AI products, open-source tools and experiments around that question.

Some work. Some fail. I share both.

Currently building/exploring:
• Reel Studio + Voiceforge
• AI + developer tooling
• QuantX Algo
• Startup experiments
• TRPCrew with my Troo mascot

GitHub ↓

[Mohit Kale @mohitkale_]

@OpenAI Agents move in minutes. Audits take months. That is not a small operational gap. Faster observability is the need of the hour to make AI safe and secured.

[Mohit Kale @mohitkale_]

Do not buy an AI plan for the feature list.

Test one job you already repeat.

Score quality, corrections, setup time, limits, and cost on the same source.

If it saves four minutes and you run it three times a month, the plan is probably a hobby.

[Mohit Kale @mohitkale_]

@benln Can we apply to multiple community groups? Or is it a hard limit to choose only 1 of them? Because it does not give an option to choose multiple options.

[Mohit Kale @mohitkale_]

The workflow that saves you hours on a repeated task.

https://t.co/Epj04kYXwq

[Mohit Kale @mohitkale_]

I built an agent workflow in 15 minutes. No code. Here's the exact walkthrough.

Every Monday I used to spend 90 minutes scanning AI & Tech news and writing it up. Not hard. Just repetitive.

So I built a three-agent workflow in n8n that drops a one-page brief in my inbox before breakfast.

Why n8n and not the others? Zapier is easiest, but it charges per task and agents get expensive fast. Make's AI Agents have been generally available since May 2026 — flexible, billed in operations. n8n is the one I used: 2.0 went stable in December 2025, AI Agent nodes are built in, and since June 2026 those agents can search the web from the Advanced panel. Self-host Community edition for free. Cloud is a 14-day trial, then paid Starter.

𝗛𝗲𝗿𝗲'𝘀 𝘁𝗵𝗲 𝗯𝘂𝗶𝗹𝗱.

𝘚𝘵𝘦𝘱 1: Schedule Trigger (2 minutes) New workflow. Add Schedule Trigger. Every Monday, 8:00 AM.

𝘚𝘵𝘦𝘱 2: Research Agent (5 minutes) Add an AI Agent node. Prompt: search the last 7 days of news, launches, and pricing for your 5 keywords. Numbered list. Source on each finding.

Model: connect a current Chat Model sub-node. GPT-5.6 or Claude Sonnet 4.5 both work.

Tools: open the agent's Advanced panel and turn on web search. Use the model's native search if it offers one. Otherwise Brave Search or SearXNG. No HTTP Request node required.

Add Window Buffer Memory, 5 messages. Hit Execute. You should get a structured list in 20 to 40 seconds.

𝘚𝘵𝘦𝘱 3: Analysis Agent (4 minutes) Second AI Agent. Prompt: identify the 3 developments that matter for your business, and what to do about each. The research output flows in automatically.

𝘚𝘵𝘦𝘱 4: Writing Agent (3 minutes) Third AI Agent. Prompt: 1-page brief. Two-sentence summary, three developments, a "What We Should Do" list.

𝘚𝘵𝘦𝘱 5: Email (1 minute) Gmail or Outlook send node. Your address. Subject with the week. Body = the brief. Activate the workflow.

That's it. Next Monday it runs without you.

Cost: a few rupees to a couple of tens of rupees in model API per run. A weekly brief typically stays well under Rs 100 a month. n8n itself is free if you self-host.

Someone will say this is just three prompts in a row. Fair. And that's the point. Research, analyse, write. Clear instructions. A review when it lands. Then you automate the chain.

The first workflow doesn't need to be impressive. It needs to be useful.

[Mohit Kale @mohitkale_]

@ericzakariasson Hey Eric, did you guys dropped the Composer model series? I mean, can we still expect Composer 3 or will it now be Grok models only?

[Mohit Kale @mohitkale_]

@LyalinDotCom It's unfortunate that the company everyone looked up at, is still way behind in terms of building a frontier LLM that beats the best out their. Come on, team Google and Gemini, bring in something crazy... we need you back into the race.

[Mohit Kale @mohitkale_]

@mastery_in_ai Yes, that's the exact reason. I prefer Human in the Loop (HITL) for curated content and mission critical workloads. Automation helps with repeating tasks, while the decision still stays in my control.

[Mohit Kale @mohitkale_]

I built an agent workflow in 15 minutes. No code. Here's the exact walkthrough.

Every Monday I used to spend 90 minutes scanning AI & Tech news and writing it up. Not hard. Just repetitive.

So I built a three-agent workflow in n8n that drops a one-page brief in my inbox before breakfast.

Why n8n and not the others? Zapier is easiest, but it charges per task and agents get expensive fast. Make's AI Agents have been generally available since May 2026 — flexible, billed in operations. n8n is the one I used: 2.0 went stable in December 2025, AI Agent nodes are built in, and since June 2026 those agents can search the web from the Advanced panel. Self-host Community edition for free. Cloud is a 14-day trial, then paid Starter.

𝗛𝗲𝗿𝗲'𝘀 𝘁𝗵𝗲 𝗯𝘂𝗶𝗹𝗱.

𝘚𝘵𝘦𝘱 1: Schedule Trigger (2 minutes) New workflow. Add Schedule Trigger. Every Monday, 8:00 AM.

𝘚𝘵𝘦𝘱 2: Research Agent (5 minutes) Add an AI Agent node. Prompt: search the last 7 days of news, launches, and pricing for your 5 keywords. Numbered list. Source on each finding.

Model: connect a current Chat Model sub-node. GPT-5.6 or Claude Sonnet 4.5 both work.

Tools: open the agent's Advanced panel and turn on web search. Use the model's native search if it offers one. Otherwise Brave Search or SearXNG. No HTTP Request node required.

Add Window Buffer Memory, 5 messages. Hit Execute. You should get a structured list in 20 to 40 seconds.

𝘚𝘵𝘦𝘱 3: Analysis Agent (4 minutes) Second AI Agent. Prompt: identify the 3 developments that matter for your business, and what to do about each. The research output flows in automatically.

𝘚𝘵𝘦𝘱 4: Writing Agent (3 minutes) Third AI Agent. Prompt: 1-page brief. Two-sentence summary, three developments, a "What We Should Do" list.

𝘚𝘵𝘦𝘱 5: Email (1 minute) Gmail or Outlook send node. Your address. Subject with the week. Body = the brief. Activate the workflow.

That's it. Next Monday it runs without you.

Cost: a few rupees to a couple of tens of rupees in model API per run. A weekly brief typically stays well under Rs 100 a month. n8n itself is free if you self-host.

Someone will say this is just three prompts in a row. Fair. And that's the point. Research, analyse, write. Clear instructions. A review when it lands. Then you automate the chain.

The first workflow doesn't need to be impressive. It needs to be useful.

[Mohit Kale @mohitkale_]

@AdryanV90 @theo Performance at the cost of battery draining faster than before? My fully charged iPhone discharges at-least 2hrs faster than before.

[Mohit Kale @mohitkale_]

@Gemini_Notebook I have been requesting this for a while, an option to select a voice. What better than supporting voice cloning?

It's a notebook, from students to engineers, everyone would love to hear their notes in their own voice. When is Gemini Notebook planning to surprise us?

[Mohit Kale @mohitkale_]

@TorrontoArts I wrote a couple of articles last week, especially on TypeSafe's Jev utilities, would appreciate your thoughts on it. I have a couple more coming this week and next week, especially my learnings -> mistakes to pivots and a few other technical project specs.

[Mohit Kale @mohitkale_]

@GoogleAIStudio I will continue building my local-first open-source REEL STUDIO project.

https://t.co/jDD423pTWj

[Mohit Kale @mohitkale_]

I have been building my local-first open-source Reel Studio project. So far, it can do the following locally,

audio TTS (multi-lingual)
podcasts (multi-voices)
animated videos
generate in 9:16, 16:9, or 1:1 MP4 formats
AI integrated automation

Checkout/fork the GitHub link below 👇

[Mohit Kale @mohitkale_]

I have been building my local-first open-source Reel Studio project. So far, it can do the following locally,

audio TTS (multi-lingual)
podcasts (multi-voices)
animated videos
generate in 9:16, 16:9, or 1:1 MP4 formats
AI integrated automation

Checkout/fork the GitHub link below 👇

[Mohit Kale @mohitkale_]

@awsdevelopers This is the same question that was asked to me, when I once got interviewed at Amazon. Now, don't ask me what happened next...

[Mohit Kale @mohitkale_]

@reach_vb Vaibhav, you should start giving us some hint, what to expect from the DevDay? Surprises feel better when you start releasing its teaser and trailer to build up the excitement.

[Mohit Kale @mohitkale_]

@TorrontoArts Good, keep it up and all the best with your plans.
Hopefully, you will have them executed the way you want.