I’m exploring a startup idea focused on improving workflows inside Canadian engineering consulting and professional services firms.
I’m looking to connect with a technical cofounder or collaborator with strong full-stack, AI, automation, or data product experience.
The goal is to build practical software for real consulting workflows, not just another generic AI tool.
DM me if this sounds interesting.
I’m exploring how to build a chatbot for internal company knowledge, focused on CVs, employee experience, project data sheets, and proposal-related information.
The challenge is scale: 5,000+ employees and a very large project history. I’m looking for a practical approach that can answer questions reliably using approved data only, ideally with sources, confidence, and no unsupported claims.
I’m not a technical person, so I’d really appreciate guidance from anyone who has built something similar, especially around Python-based search, structured databases, deterministic chatbots, or non-LLM approaches.
Any lessons learned, tools, architecture ideas, or “don’t do this” warnings would be very helpful.
@anupamrjp Building AI tools for municipal infrastructure by day (watermains, storm sewers) and siril-bxt for astrophotography by night. Quiet launches, consistent shipping 🚀
@KaiXCreator This shouldn't be unpopular. I've built AI tools for municipal infrastructure and astrophotography plugins — never owned a Mac. The tools matter more than the logo on your lid.
Just replicated an entire SaaS landing page from a video recording - pixel-perfect.
Built entirely with Kimi K2.6 in a single conversation. No IDE. No local setup. Just described the video frames and watched it generate 66KB of production-ready HTML.
a fintech dashboard landing page with:
• Animated hero dashboard mockup with live SVG charts
• Glass-morphism navigation
• 12 fully-designed sections
• Interactive accordions, pricing toggles, floating UI cards
• Responsive grid + hover micro-interactions
@mcuban Not wrong. We deployed AI agents at work to catch consistency issues and grammar mistakes in our docs.
Funny part: every time we run them, they find completely different errors.
The consistency checker has a consistency problem.
@rileybrown Thank you for posting this. I genuinely thought I was the dumbest person on the internet for the past month.
Memory fall off had me gaslighting myself into thinking I never gave it those instructions.
Is it just me or is GLM 5.1 through @ollama noticeably dumber than GLM on https://t.co/6HmmGiW1Vr?
Same model name. Completely different IQ. Like meeting your favorite author after they have had 3 beers.
Spent the last few weeks running @claudeai , Codex,@MoonshotAI_ Kimi K2.6, https://t.co/ioYy7u3nF7 GLM-5.1, and @MiniMax_AI MiniMax 2.7 side by side on real coding work.
Verdict:
• MiniMax: useless for coding. Solid for agentic tasks though.
• GLM-5.1: genuinely impressive output, killed by a tiny context window.
• Kimi K2.6: workhorse. Long runs, holds the plot.
• Codex: best for frontend.
• Claude: planning, ideas, code review.
Final workflow: Claude plans and reviews. Codex builds the frontend. Kimi handles the long-running implementation grind.
Vibe coding update: it works beautifully right up until the moment it doesn't.
One sloppy prompt today wiped out a chunk of my app that was working fine. No undo, no recovery, just "sure, let me rewrite that for you" energy from the model.
The vibes giveth and the vibes taketh away.
How many of you remember OpenAI Playground?
Simple interface. Direct control. You could actually see what the model was doing
Feels like we lost some of that while chasing bigger, flashier AI
Makes me think… the real opportunity isn’t always the next big platform
It’s small, focused tools that solve one problem really well
What’s one workflow you wish felt as simple as that again?
Everyone’s chasing “big AI ideas”
Meanwhile, the real impact is hiding in small, boring problems
The kind that waste hours every week in niche workflows
I’m building tools for those moments (starting with CAD drawings)
What’s one tiny, annoying task in your industry you wish AI could just handle?