2026-now
VitaDAO, agentic AI consultant (contract)
Zug, Switzerland (remote)
Coding and debugging agents with verification in the loop, running in sandboxed containers with persistent memory.
more
I build and maintain agents that code, debug, and check the work. The harnesses are Codex, Claude Code, DeepSeek, and Pi, with verification in the loop. I led the architecture for personal agents that run in stateless e2b containers, using OpenClaw and Nous Hermes, with qmd and Supermemory as the memory layer. I also built the tracing dashboards for monitoring and evaluation, on Prometheus and Grafana.
2025-2026
HireHawk, senior AI solutions engineer (contract)
Texas, USA (remote)
Multi-agent workflows across Zoho CRM, Workable and n8n that saved close to 48 hours of manual work a week.
more
I led the architecture and deployment of an LLM platform: backend APIs, scalable services, eval-driven CI, and cloud containers. I ran multi-agent workflows in CrewAI and connected them to Zoho CRM, n8n, Workable, and Ribbon, with observability and real-time evaluation on the internal workflows.
2025
Production RAG on Azure for a regulated asset manager, grounded with BM25 retrieval.
more
I improved RAG pipelines on Azure Semantic Kernel, in TypeScript and Python, keeping command and query paths separate. I grounded answers with BM25 in Azure AI Search so responses stayed on the documents.
2022-2024
CVC Network, senior software engineer, AI and ML
Bangalore, India
Led a team of 5 serving translation and text-to-speech models for 400+ languages in production.
more
I led the team that served translation and text-to-speech models for 400+ languages, on FastAPI, Docker, Kubernetes, and Baseten. I set up the organisation-wide MLOps pipeline and an AI governance framework used across 22 offices. Falcon and Llama 2 ran in environments that had to stay inside GDPR. I used LangChain, LangSmith, vector databases, and embedding models to get experiments and client demos out faster. I also shipped a generative content product, GPT-4, DALL-E, and avatars, that took a multi-week production cycle down to a few hours. Two of five daily challenges at the Intel and Prediction Guard Advent of GenAI hackathon, 2023, landed in the top entries. I coordinated the GenAI roadmap with stakeholders in Australia and the Americas, and did R&D on digital twins used as chat agents.
2021-2022
Pipelines over 800+ data sources in 35 countries, with daily reporting to the board.
more
I built and maintained pipelines for ingest, extraction, cleaning, and preparation, feeding SQL reporting and the warehouse. I governed a catalogue of 800+ social sources across 35 countries, including Facebook, YouTube, Google, TikTok, Instagram, Yandex, and FrontApp. Daily reports went to the founders' office, the board, and trustees.
2026
claude-guardrails is a set of safety hooks for Claude Code. It stops destructive API calls, force pushes, and secret files before they run.
2026
nudge pings you on macOS when Claude Code finishes or is waiting, and only if you are not already looking at that terminal.
2025
mlops stands up a KinD cluster from a GitHub Action, checks the services, load-tests them with k6, and posts the result on the pull request.
2023
Advent of GenAI, with Intel and Prediction Guard. Two of the five daily challenges landed in the top entries. One was a storytelling bot on Zephyr-7B that stayed in the story and kept building on earlier replies. The other was a Stable Diffusion flipbook, Adrift at dawn.
talk
Accelerating content creation, translation and text-to-speech, from days to seconds.
2024
HLRS is the High-Performance Computing Center Stuttgart, Germany's national supercomputing center.
more
The course ran on 17 October 2024. 50 hours of lectures and exercises, covering data analysis and AI on an HPC cluster, AI ethics, exploratory data analysis, machine learning, neural networks, deep learning with PyTorch, and big-data frameworks. Also a capstone on analysing data with large language models, plus pre-processing, feature engineering, regression, classification, and Apache Spark from Jupyter on the cluster. Taught by Dr.-Ing. Lorenzo Zanon and Junghwa Lee.
2025
The Vienna Scientific Cluster is Austria's shared supercomputer for its universities.
more
The course focused on profiling AI and machine learning so it actually uses the machine: NVIDIA Nsight Systems, GPU profiling, distributed data parallelism, model parallelism, and multi-GPU and multi-node runs. Labs were on the Leonardo booster partition, NVIDIA A100s. Taught by Tosin Adesuyi from NVIDIA, with the EuroCC centres. Online, 4.5 hours.
2021
Data for Better Lives: A New Social Contract, World Bank Group
more
How public and private institutions can use data to improve lives. Taught by economists and scientists from the World Bank, on the World Bank Online Learning Campus.
2020
AWS Technical Professional (Digital)
more
The digital exam on AWS architecture, security, and deployment. Issued 27 January 2020.
2020
LTI Certified Cloud Professional
more
ITIL v3, Windows Server 2016, RHEL 7, Cisco routing and switching, and MSSQL 2017. Issued 10 January 2020.