Umair A.

Sometimes you don't need another developer. You need a real engineer's and architect's eyes on your system. I take AI and software from business problem to production, and I fix what's slow, fragile or expensive along the way: AI systems (RAG, fine-tuning, agents), cloud and server costs, and telephony/VoIP. I'm Head of Software Architecture at Voxbi, a cloud telephony + AI platform in Luxembourg used daily by enterprise customers. 12+ years building software, leading a team of 9+ engineers. 1) AI FROM IDEA TO PRODUCTION • Recommend the right approach: RAG, fine-tuning or a hybrid, with honest trade-offs on cost, accuracy and maintenance • Data preparation: collect, clean, consolidate and label your data; build training and evaluation datasets for your business • Fine-tune open models (Llama, Qwen, Mistral, Whisper) with LoRA/QLoRA, and build RAG pipelines (embeddings, hybrid search, re-ranking) • LLM agents with tool use and MCP, plus evaluation, guardrails and monitoring • Integrate it all into your apps, CRM/ERP (HubSpot, Salesforce, Dynamics, Odoo) and phone system 2) CLOUD COST & PERFORMANCE OPTIMIZATION Track record: up to 80% lower server costs on the same code and the same servers, with the same or better performance. • AWS, Google Cloud and Azure: cost audits, right-sizing, architecture review • PHP/Laravel performance: PHP-FPM tuning, OPcache, slow queries, N+1 problems, caching • Redis, Laravel Horizon, Supervisor and queue workers: memory leaks, stuck jobs, crashes • Nginx, MySQL/PostgreSQL tuning, Docker, CI/CD, Linux server setup and hardening • Grafana monitoring and dashboards, so issues are caught before your customers notice 3) TELEPHONY, PBX & VOICE AI • Asterisk/FreePBX PBX setup and deployments, built from scratch or rescued • SIP trunks, third-party telephony integrations, WebRTC softphones, CRM click-to-call • Common PBX issues fixed: registration and connection drops, one-way audio, NAT, codecs, VAD/silence detection • Voice AI: transcription, call summaries, live call assist, AI voice agents over SIP HOW WE WORK TOGETHER 1. Understand your business, your system and your constraints (budget, privacy, latency) 2. Audit and recommend: a written plan with cost, risk and expected impact 3. Build and fix: data, models, code, servers and integrations 4. Measure: before/after numbers, evaluations and Grafana dashboards 5. Ship to production and keep improving RECENT WORK • Fine-tuned local telephony LLMs and built production RAG on the MeluXina EuroHPC supercomputer, fed by a centralized, fully automated data pipeline that turns any business's data into RAG knowledge and fine-tuning datasets • Designed a browser-local AI chat for a telephony platform (Qwen on WebGPU, SQLite-WASM, MCP), so customer data never leaves the device • Fine-tuned a Luxembourgish speech model for live calls, including the full dataset pipeline • Built CRM telephony connectors (HubSpot Calling SDK, Dynamics CIF) for Benelux/DACH businesses • 14 live Grafana dashboards turning call records and AI call analysis into business insights WHY WORK WITH ME One accountable engineer from the first call to production: architect, AI engineer, DevOps and telephony expert in one. No hand-offs between specialists. EU-based (CET), GDPR-aware, fluent English. BSc Computer Science. Tell me what you're trying to solve, or what's costing you too much. Within 24 hours I'll reply with a recommended approach and a rough plan.