This week I'm excited to welcome back Sreemanti Dey. She's an ML engineer at Abacus.AI, with a focus on generative AI, LLM fine-tuning, and full-stack development.
In just 4 minutes, Sreemanti breaks down the ML project that she used to turn interviews into offers. (She'll also share 3 disqualifying mistakes most ML portfolios make, so look out for that, as well).
👉 Last thing: Halloween offers are live on Educative now. That means it's a great time to grab epic deals on Premium and Premium Plus subscriptions, so you can access all the job-ready projects you need to perfect your portfolio. Use this link to see your deals.
Now over to Sreemanti!
Hey everyone đź‘‹, Sreemanti here.
If you’re tired of polishing tutorials and hearing crickets from recruiters, this is your wake-up call.
In 2025–2026, AI/ML roles are exploding, but portfolios full of Kaggle clones won’t cut it.
You need one thoughtful, end-to-end project that shows judgment, impact, and the ability to ship.
One Project That Got Me Hired in AI - Steal This 4-Minute Playbook
▶️ Watch the video:How I Built A Machine Learning Project That Got Me Hired
Quick note before we dive in: my unfair advantage for both ideas and execution was Educative. For project-driven learning—agentic AI, LLMs, GenAI, NLP, ML, and system design—it’s practical and lets you ship. I’ll reference it throughout, so you can jump straight into building.
❌ Same Kaggle clones with no originality ❌ Missing end-to-end ownership (API, UI, deployment, observability) ❌ Accuracy ≠Impact (user outcomes are what matter!)
Hiring managers don’t care about what cross-entropy is—they care about how you kept p95 latency under 300ms.
🌱 The Project That Flipped My Job Search: Plant Doctor
Plant Doctor is a clean, fast web app that lets users upload a plant photo to get:
An AI-backed diagnosis of issues (pests, nutrients, diseases)
A care plan with clear remediation steps
Ongoing reminders and a history of past diagnoses
Multi-plant tracking and “buy treatment” links
What made it special? Not just the ML model, but the system:
Backend/data modelling that’s boring (in a good way) and reliable
I go deeper into the stack, tradeoffs, and the exact architecture in the video. Watch it to see how the pieces fit and why the final product stood out.
đź’¬ What Interviewers Actually Ask
Latency and cost: How did you keep it stable and affordable?
Evals and quality: How do you measure the AI component beyond accuracy?
Monitoring: What signals tell you the system is drifting or failing?
Tradeoffs: Why this stack, this cache, this deployment setup?
If your project invites these questions, you’re no longer “another resume.”
You’re the candidate with proof.
đź’ˇ Two More Portfolio Winners You Can Ship
Mood Tunes: Detect a user’s mood from a selfie and auto-curate playlists. Great for multimodal inference + third-party APIs + product polish.
Art Identifier: Snap a painting, get its backstory, and surface similar works. Perfect for retrieval + generation + delightful UX.
Keep them small, polished, and measurable, then iterate.
🧠The Execution Mindset Wins
âś… Write a 1-page PRD (problem, users, constraints, success metrics) âś… Ship a tiny version fast (one flow, one model path, one UI) âś… Track outcomes (latency, $/inference, satisfaction) âś… Document tradeoffs (X vs Y decisions)
This is how you move from “I learned ML” → “I can build AI products.”
🔑 Ready to Build Your Own High-Signal Project?
My secret weapon was Educative, a goldmine for project-based learning:
Agentic AI
LLM apps
MLOps
System Design
If you want to follow my exact path, start here (which is available to you at a big discount if you use this link): 👉 Hands-On Projects
Now go build the thing that makes recruiters hit “reply.”
If you enjoyed this, you can always connect with me on Topmate or check out my videos on YouTube.
