Summary

Overview

Work History

Education

Skills

Work Preference

Timeline

MUHAMMAD IBRAHIM

Artificial Intelligence | Backend Developer

Lahore,PB

Summary

Results-driven Artificial Intelligence Engineer with expertise in applied machine learning and deep learning, focusing on Natural Language Processing, Large Language Models, Speech AI, and Generative AI orchestration. Delivered impactful AI solutions through advanced prompt engineering and state-of-the-art model fine-tuning, optimizing performance and enhancing team collaboration. Committed to developing scalable, efficient AI algorithms and production-ready applications across cloud and edge-compute environments.

Overview

Languages

years of professional experience

Work History

AI Engineer - Speech and Generative AI

TalkIt

06.2024 - Current

Engineered a real-time speech transcription pipeline using Soniox and Deepgram SDKs, integrated via WebSocket connections for live consultation audio streaming with sub-second latency.
Leveraged advanced Soniox features including speaker diarization (multi-speaker labeling), confidential information redaction, and noise filtering to ensure privacy-compliant medical transcriptions.
Implemented intelligent role detection using GPT-4.1 to automatically classify speakers as Doctor or Patient in real time, enabling context-aware downstream processing of consultation dialogues.
Built and maintained WebSocket server endpoints to handle concurrent real-time audio streams, ensuring fault-tolerant and scalable transcription infrastructure.
Designed and iterated multi-model prompt pipelines (GPT-4.1, GPT-5.4, Claude Sonnet 4.6) to generate structured clinical documentation from transcription outputs.
Engineered prompts for SOAP Notes (Subjective, Objective, Assessment, Plan), Sick Leave certificates, Consultation Summary reports, Referral Letters, and Magic Edit (free-form note refinement).
Applied chain-of-thought and role-system prompt techniques to improve factual accuracy, clinical terminology adherence, and output consistency across diverse consultation types.
Validated and benchmarked medical note quality against clinical documentation standards, iterating prompts to reduce hallucination rates and improve structured field extraction.
Prompt Engineering - Hospitality Domain
Developed prompt templates for hotel management use cases including Meeting Minutes generation, Yearly Employee Assessment Notes, and operational summaries using GPT-4.1, GPT-5.4, and Claude Sonnet 4.6.

ASR and TTS Engineer

Invictus Solutions | Elite IT | 7 Kings Code

03.2023 - 05.2024

Developed and deployed production-ready Automatic Speech Recognition (ASR) and Text-to-Speech (TTS) pipelines for call center voice-assistant systems, improving transcription accuracy across multiple languages.
Implemented the Whisper ASR model and Google Speech-to-Text API for high-accuracy audio transcription across multiple languages and accents.
Engineered a voice transformation pipeline using the Coqui TTS architecture, enabling custom voice synthesis for conversational agents.
Integrated ElevenLabs TTS API into a callbot system, delivering natural-sounding AI-generated responses within real-time dialogue flows.
Built and tested end-to-end voice interaction pipelines covering audio capture, ASR, NLU, response generation, and TTS playback.
Fine-tuned multiple open-source Large Language Models (LLMs) including Llama 3, Mistral 7B, and Zephyr 7B for domain-specific applications using QLoRA with 4-bit and 2-bit quantization on resource-constrained hardware.
Synthesized and curated training datasets in CSV/JSONL formats, applying data cleaning, deduplication, and instruction-tuning formatting pipelines for optimal fine-tuning outcomes.
Applied Retrieval-Augmented Generation (RAG) with LangChain and OpenAI APIs to deliver accurate, citation-grounded responses for knowledge-intensive applications.
Built a real-time Stocks Chatbot generating live financial data summaries and analysis using integrated financial APIs and prompt-engineered LLM pipelines.

Computer Vision Engineer

Invictus Solutions | 7 Kings Code

01.2022 - 04.2022

Designed and deployed real-time crowd counting and people detection systems using YOLO object detection models, achieving high accuracy in dense and complex scenes.
Optimized YOLO model architectures (YOLOv5, YOLOv7, YOLOv8) for deployment on Raspberry Pi edge devices, achieving real-time inference on resource-constrained hardware by reducing model size while maintaining detection accuracy.
Implemented multi-camera real-time surveillance systems for intrusion and theft detection, enabling concurrent processing of video streams with minimal latency.
Developed document intelligence pipeline using YOLOv8 for automated resume field extraction, efficiently identifying and parsing structured data from unstructured PDF and image documents.
Implemented Chain-of-Thought (CoT) prompting strategies and integrated DeepSeek-Coder for structured multi-step code generation and reasoning tasks.
Integrated open-source models (Open Orca, Flan-T5) as Bittensor subnet miners, enhancing competition against validators in decentralized AI incentive network.

Education

Bachelor of Science - Software Engineering

Riphah International University

Faisalabad, PB

01-2023

Higher Secondary Certificate - FSc Pre-Engineering

Govt. Municipal Degree College

Faisalabad, PB

01-2018

Skills

Speech AI: Soniox, Deepgram, Whisper ASR, Google Speech-to-Text, Coqui TTS, ElevenLabs TTS

Generative AI models

Prompt engineering techniques

AI/ML Frameworks: TensorFlow, PyTorch, Scikit-Learn, Keras, Hugging Face Transformers

Statistical Modeling

Model evaluation

Programming Languages: Python (Expert), R, HTML, JavaScript

PyTorch framework

Speech synthesis

Agentic AI orchestration

Python