I write code. I make it work. I make it stay working.

$ whoami
> software engineer & applied ai.
> currently: shipping small models, building production systems.
> stack: python, fastapi, pytorch, next.js, postgres.
> learning: applied ai & llms.

find me: github · huggingface · linkedin · email

BlitzKode | Open Source LLM 4-stage fine-tuning pipeline (SFT -> Reward-SFT -> DPO -> LoRA) on Qwen2.5-1.5B-Instruct. VRAM from 24GB to 8GB. PyTorch + PEFT training, exported to GGUF for CPU-only inference via llama.cpp. 510+ downloads on HuggingFace. [PyTorch] [PEFT] [llama.cpp] [FastAPI]

EsenceLab | AI Hiring Platform 3-service monorepo. Groq-powered parsing + Supabase RLS for strict row-level security. Won 2nd prize at Innovision (out of 30 teams). Zero downtime in production. [Next.js] [FastAPI] [Supabase] [Docker]

Lexora AI | Multi-Tenant RAG Document Q&A with per-user FAISS index isolation. Celery for background chunking and Redis for caching. Sub-2s latency on 100k+ token corpora. [LangChain] [FAISS] [Celery] [Redis]

SentinelML | Fraud Detection API RandomForest with SMOTE oversampling. Redis prediction cache to handle repeated requests. Drift tracking with MLflow. 0.94 ROC-AUC. [Scikit-Learn] [MLflow] [PostgreSQL]

AI Resume Screener | Resume-JD Matcher FastAPI-based resume screener using Groq (llama-3.1-8b-instant). JWT auth, screening history, admin role, statistics dashboard. SQLite + SQLAlchemy. [FastAPI] [Groq] [SQLAlchemy] [JWT]

SpeakSwap | Translation App Next.js 15 + React 19 web app. Text + speech translation between languages via free endpoints (MyMemory, LibreTranslate, Lingva) with fallback. Voice input/output, dark mode, translation history in localStorage. [Next.js] [React] [TypeScript] [Tailwind]