Open to opportunities

Amish Bemelkhedkar

@amishbemelkhedkar

Message

AI/ML engineer fine-tuning 7B LLMs end to end, fixing collapsed training runs.

India

Message

What I'm looking for

I’m looking for a remote ML role where I can build and evaluate ML/LLM training pipelines end to end—dataset work, fine-tuning, and sandboxed regression-safe evaluation.

I’ve taken 7B-parameter LLMs end to end—built the dataset, ran 4-bit QLoRA fine-tunes on GPU, diagnosed why accuracy collapsed to 2.47%, and recovered training to beat the baseline.

In my MTech thesis at NIT Goa, I fine-tuned DeepSeek-Coder-6.7B and Qwen2.5-Coder-7B to enforce functional code generation. I root-caused a failed functional-only run and brought it back using a multi-paradigm curriculum interleaving functional, imperative, and hybrid data.

I’m comfortable owning the boring-but-critical parts: dataset design from scratch, ingestion, filtering, deduplication, and QA with unit-test-based verification. I also build sandboxed automated eval harnesses and track checkpoint regressions.

Outside LLM work, I trained a Random Forest for smart fertilizer recommendations from noisy sensor telemetry, and shipped web/backend systems like an institutional MIS portal with RBAC/SSO and a digitised criminal registration workflow with MySQL schemas.

Experience

Work history, roles, and key accomplishments

Fine-Tuning LLMs for Functional Code Generation

National Institute of Technology Goa

Jan 2025 - Jan 2026 (1 year)

Fine-tuned DeepSeek-Coder-6.7B and Qwen2.5-Coder-7B with 4-bit QLoRA to enforce functional programming paradigm. Root-caused a failed run and recovered performance with multi-paradigm curriculum learning.

PyTorch QLoRA Functional Programming Curriculum Catastrophic Forgetting Unit Testing

Smart Fertilizer Recommendation

National Institute of Technology Goa

Trained a Random Forest model for per-field nutrient recommendations from noisy time-series sensor telemetry. Owned data cleaning, anomaly detection, and outlier handling.

Random Forest Machine Learning Time Series Anomaly Detection Data Cleaning

AI/ML Engineer

Self Employed

AI/ML engineer with MTech CSE from NIT Goa, experienced in fine-tuning LLMs, building datasets, and evaluation harnesses.

Criminal Registration System

National Institute of Technology Goa

Digitised a paper-based police complaint workflow into a centralised system. Designed normalised MySQL schemas for reliable storage and fast retrieval.

MySQL Database Design Backend Development

MIS Portal — Teacher & Administrator Module

National Institute of Technology Goa

Shipped a web portal for institutional data management with Role-Based Access Control, Single Sign-On, and HTTPS. Gathered requirements from faculty users.

RBAC SSO HTTPS Web Development Full Stack