Hammad Safi.
I build reliable AI systems.
I'm a software engineer with an MSc in Advanced Computer Science from the University of Hertfordshire, focused on Python, backend APIs, and AI-powered systems. I build production-grade RAG pipelines, multi-turn agentic applications, and LLM fine-tuning workflows — software that holds up under real traffic and solves real problems.
01. About Me
I enjoy taking fuzzy problems — ambiguous user intent, messy data, tight latency budgets — and turning them into systems that are predictable, observable, and safe to ship. My work sits between research and production: I've fine-tuned open-weight LLMs on domain tasks, built RAG pipelines that cut token costs without sacrificing answer quality, and designed evaluation loops that track semantic correctness beyond BLEU scores.
Recently I've been focused on agentic systems — state machines, fallback rules, and tool-calling paths that don't unravel when users don't behave like the training data. Outside of code, I led a 500+ member engineering society at university and placed as a finalist at the Smart India Hackathon.
Technologies I've been working with:
- Python (FastAPI, Flask)
- PyTorch & HuggingFace
- LangGraph & Pydantic AI
- RAG · Qdrant · vLLM
- LLM fine-tuning & distillation
- DistilBERT & Transformers
- REST APIs & Microservices
- SQL & Firebase
- Docker · Linux · Git · TDD
02. Where I've Worked
AI Systems & Software Developer · Freelance
- Led architecture for multi-turn agentic applications, designing state management frameworks and fallback rules to handle unpredictable real-world user intent.
- Engineered production-grade RAG pipelines in Python with FastAPI and vector databases; optimised retrieval to reduce token inference costs and end-to-end latency.
- Fine-tuned and distilled open-weight LLMs via PyTorch and HuggingFace for domain-specific intent extraction and deterministic tool-calling.
- Built automated evaluation loops tracking semantic correctness, classification errors, and user corrections to keep deployments safe in production.
Customer Operations · Ladbrokes / Entain
- Work within high-volume customer channels requiring real-time exception handling, complex problem-solving, and adherence to rigorous procedures.
- Resolve ambiguous operational conflicts safely and efficiently under time-sensitive constraints.
Web Engineering Lead · Amity University Gurugram
- Directed cross-functional teams to spec, build, and deploy production-grade responsive web applications.
- Architected and shipped the Amifest platform, supporting 6,000+ unique views within 24 hours without service degradation.
- Mentored junior engineers on code quality, testing practices, and Git workflows; authored reusable deployment templates.
03. Things I've Built
Python application that analyses prompts across goal, audience, format, constraints and context, classifies missing information via a DistilBERT multi-label model, and reconstructs improved prompts through LLM-backed workflows. FastAPI endpoints and SQLite-backed clarification flows.
Deep-learning sequence modeling (LSTM) compared against classical baselines (Random Forest, ARIMA) on multivariate time-series weather data, with an observable Streamlit dashboard tracking model performance and live evaluation.
Real-time computer vision pipeline using YOLOv9 and TensorFlow for drowning detection on live video streams. Data augmentation and network pruning keep inference at 32fps with 89% mAP@0.5 for edge deployment.
Get in touch.
I'm currently open to AI/software engineering roles and agentic-systems collaborations. Whether you have a role, a project, or just want to talk LLMs and Python — my inbox is open.
Say Hello