Manindra de Mel Canberra  AEDT

Manindra de Mel Software engineer & ML researcher

©2019–2026

Corresponding author
PlantCLEF 2026
7th on the official leaderboard

Featured work

(06)
  1. 2026Cog & LeverAutomation & ML
  2. 2026Plan4MeAI travel app
  3. 2026PlantCLEF 2026ML research
  4. 2026chr0me.au3D e-commerce
  5. 2023–Rocket RemitFintech data
  6. 2023Canberra Cloud KitchenOrdering platform
  7. All work

Research

(03)

Fine-Tuning of BioCLIP 2.5 with Taxonomic Heads for Multi-Species Plant Identification

Arjun Raj, Manindra de Mel, Razeen Wasif, William Brake

A PlantCLEF 2026 test quadrat: rock, bare soil and mixed low vegetation inside a sampling frame
  1. Quadrat: one field photo of a sampling frame
  2. Sixteen tiles: the photo is cut into a 4 by 4 grid
  3. Two resolutions: each tile is encoded at 224 and 336 pixels
  4. One species set: predictions are averaged, adjusted and thresholded
0.418Macro F1
7thLeaderboard
7,806Species
2.65MTraining images
Paper ↗ Project site ↗ Code ↗

More papers

About Manindra de Mel

I build software that leaves the lab. That means AI products people carry in their pockets, vision models trained on messy field data, and back-end systems that keep payments moving. I work small, ship end to end, and stay on call after launch.

Products

Apps taken from sketch to store, with live data, streaming generation, payments and subscriptions.

Models

Fine-tuning and evaluation for vision models that have to work on real field data.

Systems

Payments data pipelines, ordering platforms and analytics, run in production.

Manindra de Mel standing beside a car at a petrol station at night

Most projects reach me in one of three shapes. A product that has to go from sketch to App Store. A model that has to hold up on real-world data. A system that has to stay up while the business grows around it. I scope it, build it and run it, usually as the only engineer.

Cog & Lever↗ Résumé↗ GitHub↗

Kind words

(04)
  1. Rocket Remit
    “Strong technical work on our data pipeline and reporting. Manindra understood the regulatory context quickly, asked the right questions, and produced output we could hand straight to the team. Considered, dependable, and easy to work with.”
  2. Canberra Cloud Kitchen
    “Manindra delivered a clean, reliable ordering platform that fit the way our kitchen actually works. He scoped the brief carefully, kept us in the loop through every iteration, and shipped on time. Order flow has been smooth since launch and the back-of-house team picked it up immediately.”
  3. Quintet Automotive
    “Excellent service in data analysis and web scraping for our automotive business. Deep understanding of analytics, ability to extract and process the relevant information, and clear communication throughout. Strongly recommend.”
  4. Cog & Lever
    “We automate the admin that eats your week, and train machine learning models on your own data for the decisions that matter.”

Extras

(11)

Field notes

  • Lessons building an AI travel plannerIn progress
  • Fine-tuning BioCLIP 2.5 for PlantCLEF 2026In progress
  • Next write-upDraft

Shipped sites

Documents