Projects · Model layer

AI Projects

Five projects: a live career product, two LoRA fine-tunes, a data-collection tool and an account decision brief. The three notebook projects run from a single Colab notebook.

Live product

HireOnAI

A career product that matches a résumé to real jobs, explains the score, names the gaps, and tailors the application without inventing experience.

Stack · Product direction · LLM APIs · Python / FastAPI

01 · Fine-tune

Polar Orca

Formerly IRA Dolphine · Built by Rajiv Buda

A LoRA fine-tune of a 4-bit Qwen2.5-3B-Instruct base with Unsloth, trained on a slice of the FineTome instruction set, with before/after checks and Hub publishing. Apache-2.0.

Stack · Python · Unsloth · TRL · PEFT/LoRA · Hugging Face

02 · Custom assistant

Fishing Cat AI

A small custom assistant named after the fishing cat, Odisha's state animal. Llama-3.2-1B-Instruct in 4-bit, LoRA-trained on a hand-written set of 219 instruction/answer pairs.

Stack · Python · Unsloth · Llama-3.2-1B · TRL SFTTrainer · Google Drive

03 · Data utility

VK Public Posts Fetcher

A data-collection utility that pulls recent public wall posts from a VK community through the official API, with the token kept in a Colab secret.

Stack · Python · vk_api · requests · BeautifulSoup

Decision brief

Account Decision Brief

An account pack goes in. The brief comes out with a decision, sourced evidence, and a refusal to invent a number.