← Home

Work

Four projects that show how I work: a latency-bound voice programme run in production, a planner that has to keep its measurements honest, a local-first code mapper, and a fine-tuned model behind a narrow interface.

01

Conversational AI Voice Agents

Forward-deployed across four production LLM voice-agent launches — solar, real estate, fitness and restaurant ordering — covering campaign setup, CRM integration, UAT and post-launch reliability.

Client work · 2026 · Python, FastAPI, AWS Lambda, Conductor

02

GeoLLM

Ask a geospatial question in plain language. A planner selects the right tools, grounds the answer in retrieved context, and returns an analysed map instead of a plausible sentence.

Personal project · 2025 · Python, FastAPI, Google Earth Engine, FAISS

03

tldrepo

A CLI that scans a repository, draws a compact ASCII map, and puts one short, truthful description beside every file and directory — deterministic Python first, a small local model only where prose is worth generating.

Open-source CLI · 2026 · Python, Ollama, Rich, AST

04

Prompt2Shell

A terminal-style interface that turns a plain-English request into an executable shell command, served by a Phi-3-mini model fine-tuned with QLoRA.

Personal project · 2025 · Python, QLoRA, Phi-3-mini, FastAPI