Rakib J.

I'm a senior AI engineer with 8+ years of production software experience. I build AI systems end to end, from model and pipeline design through to deployed backends that teams can actually maintain and build on. On the AI side, I work across RAG pipelines, multi-agent orchestration, LLM fine-tuning, and evaluation frameworks. That means chunking strategies, retrieval and re-ranking, agent tool use with proper failure recovery, LoRA/QLoRA fine-tuning with evaluation against baseline, and dual-LLM quality scoring pipelines that catch hallucinations before they reach users. On the backend side, I build the infrastructure AI products actually run on. FastAPI services with auth, async task queues, rate limiting, and clean data models. I've shipped to Supabase, PostgreSQL, and NeonDB, integrated vector databases for semantic search, and built Streamlit dashboards for internal tooling. If you're building an AI SaaS product and need someone who can own the Python backend — not just the LLM calls — that's squarely in my wheelhouse. For connecting AI outputs to real business systems, I use n8n and Make com. I also work with Flowise and AutoGen for teams who want lower-code agent interfaces sitting on top of solid backend logic. Shipped work → Full-stack development of BandLift, an AI Speech Analysis powered SaaS that helps users improve their IELTS speaking score . → Speech Analysis API, live on RapidAPI. IELTS-aligned fluency and pronunciation scoring from raw audio — custom Whisper and Wav2Vec stack, not a third-party wrapper. → Medical Reports AI: LangGraph pipeline for structured field extraction from unstructured medical PDFs using multi-node agent routing. → AI Book Translator: Claude API generates, OpenAI evaluates — automated quality scoring with human-readable diff output. → AI Life Companion: conversational system with persistent long-term memory and personality layers, built on Flowise and NeonDB. → FDG '26 submission: regression analysis across 35 mobile game prototypes identifying behavioral predictors of Day-1 retention. Session frequency returned r=0.84 as primary driver. Background Before AI became my primary focus, I spent eight years as a senior Unity engineer shipping 50+ mobile titles with publishers including Voodoo, Homa, and Supersonic. That background shows up in how I write code: structured, testable, and built to survive a handoff. If you're building something that needs real engineering underneath the AI — not a prototype that falls apart in week two — let's talk.