Looking for a job

Danial Rehman

@danialrehman1

Message

AI/ML Engineer at Tensor Labs who cut FindSocial API p95 latency by 82% and ships LLM systems to production.

Pakistan

Message

What I'm looking for

I'm open to full-time or contract (B2B) roles, with full EU overlap and at least four hours of US-East overlap.

At Tensor Labs, I rebuilt FindSocial’s MongoDB read path, cutting single-query latency from 20–90 seconds to about one second. In load tests, I also reduced p95 latency at 20 concurrent requests from 17.9 seconds to 1.9 seconds while increasing throughput from 1.3 to 18 requests per second.

I’ve shipped agent and RAG pipelines at Tensor Labs, and built production RAG systems at CortechSols Pvt Ltd with LangChain and vector databases. At CortechSols, I fine-tuned Llama 3 8B with LoRA/QLoRA and 4-bit quantization, reaching 89.5% accuracy on domain-specific support tickets.

Experience

Work history, roles, and key accomplishments

Current

Machine Learning Engineer

Current

Tensor Labs

Jan 2025 - Present (1 year 10 months)

Rebuilt FindSocial's MongoDB read path, cutting single-query latency from 20-90s to ~1s and improving p95 latency 9x under load. Built JARVIS, a real-time voice assistant on async FastAPI with WebSocket audio streaming, and deployed agent and RAG pipelines on AWS/Azure/GCP.

MongoDB fastAPI WebSockets Crew AI Google ADK Pinecone Qdrant ChromaDB AWS Azure GCP Docker GitHub Actions Python AsyncIO OpenWakeWord Onnx OpenAI GPT Realtime

AI Engineer

Codistan Ventures

Aug 2024 - Jan 2025 (5 months)

Built real-time speech chatbot backends in Python, integrating LLMs with dynamic context retrieval. Developed computer-vision data pipelines with OpenCV and PyTorch, and automated creative workflows with Stable Diffusion and ComfyUI.

Python LLM OpenCV PyTorch Stable Diffusion ComfyUI

AI Engineer

CortechSols Pvt Ltd

Aug 2023 - Aug 2024 (1 year)

Shipped 4-5 production RAG systems with LangChain, connecting OpenAI and Hugging Face models to Chroma, Pinecone and Qdrant. Fine-tuned Llama 3 8B with LoRA/QLoRA and 4-bit quantization, reaching 89.5% accuracy on domain-specific support tickets.