Lahore, Pakistan · Full-Stack AI Engineer @ Sensus DWC LLC (Dubai)
I'm an AI Engineer with 2.5 years of focused experience building RAG systems, LLM applications, and Voice AI agents for real clients across automotive, education, defense, and enterprise SaaS.
My work bridges research and deployment — I build things that run in production, not just demos. I specialize in:
- 🧠 RAG & LLM Systems — semantic retrieval, hybrid search, multi-tenant architectures, prompt engineering at scale
- 🎙 Voice AI Agents — real-time conversational agents using ElevenLabs, VAPI, Retell AI, Deepgram, and Twilio
- ⚙️ Backend & APIs — FastAPI, Django REST, Node.js, AWS, Docker
Open to AI/ML roles and research collaborations.
|
Client: Proton Malaysia 
|
Podcast-native RAG platform for military intelligence content. Transcribes, embeds, and semantically searches 100+ hours of military podcasts. Built with Deepgram, Pinecone, and LLMs — enabling sub-second semantic discovery across a large audio content library. 
|
|
Outbound trade-in campaign agent for an automotive dealership. Runs full conversation flows — qualification, objection handling, appointment booking — over WhatsApp via a multi-agent architecture with session memory, pgvector semantic search, and campaign-specific guardrails. 
|
🛠 AureliaAI diagnostics assistant for an industrial crane services company. Processes error codes and troubleshooting manuals for EFFER, PM, and HIAB equipment — delivering concise, accurate guidance in under 500 chars for field technicians under time pressure.
|
🧠 Ekko AIEnterprise multi-tenant SaaS for AI-assisted knowledge work. Features subscription management, analytics dashboard, prompt template library, and Firebase push notifications. Built on Django REST + LangChain — designed for team-scale deployment.
|
📘 MathWizAdaptive EdTech platform that generates grade-appropriate math questions, runs personalized quizzes, and gamifies progress with badges. LLM-generated content served via FastAPI + AWS — designed to reduce teacher workload for content creation.
|
|
Production-ready FastAPI wrapper for robust playing card detection using YOLOv8 with multi-scale detection, image preprocessing, and batch processing capabilities. 
|
Islamic AI assistant grounded in Quran & Hadith for contextual Q&A. A RAG-based system delivering domain-accurate, citation-aware religious guidance.
|
| Project | What it does |
|---|---|
| 🎓 Akademus AI | Document-based RAG platform with auto-generated flashcards & quizzes for students |
| 🧾 Requirement GPT | LLM-powered structured SRS document generator for software teams |
| 🔊 AI Voice API Platform | Omnichannel conversational AI system with automated telephony infrastructure |
AI / LLM
Backend & Infra
Databases & Vector Stores
Voice & Audio
Dev Tools
- 🥇 Kaggle Notebooks Master — Top 150 globally · 1x Gold, 5x Silver, 15x Bronze medals
- 🌍 LinkedIn Top Voice (AI) — Recognized for making complex AI concepts accessible
- 💻 250+ LeetCode Problems — Strong DSA foundation
- 🎖 Top Performer @ Knowledge Streams — 94% evaluation score during AI internship
- 🎓 IBM Certified Data Scientist
- 🛠 Building: Outbound Voice AI calling agents & multi-agent WhatsApp systems
- 📚 Reading: AI Engineering by Chip Huyen
- 💼 Open to: AI/ML engineering roles · Research collaborations
18 merged PRs across 7 major AI/ML frameworks — 2026
| Repository | PR | Merged | What was fixed |
|---|---|---|---|
| pydantic/pydantic-ai | #6113 | Jun 30 | Guard TogetherProvider.model_profile against model names without a / |
| pydantic/pydantic-ai | #6111 | Jul 3 | Align LLMJudge._build_prompt section order with its few-shot examples |
| pipecat-ai/pipecat | #4718 | Jun 17 | Add missing defaults for NvidiaSegmentedSTTService diarization settings |
| pipecat-ai/pipecat | #4840 | Jun 23 | Fix WorkerBus dropping future messages to a subscriber after a hook raises |
| pipecat-ai/pipecat | #4843 | Jun 23 | Fix Mem0MemoryService injecting empty memory prompt when none found |
| mem0ai/mem0 | #5717 | Jun 24 | Fix silent reranking failures across all 5 reranker providers |
| mem0ai/mem0 | #5715 | Jun 25 | Fix HuggingFace reranker scoring — sigmoid instead of min-max normalization |
| mem0ai/mem0 | #5960 | Jun 29 | Fix FAISS cosine search — normalize vectors for IndexFlatIP |
| infiniflow/ragflow | #16330 | Jun 25 | Fix PubMed tool author parsing (no longer returns "Unknown Authors") |
| infiniflow/ragflow | #16332 | Jun 25 | Fix DeepL component validation and silent error dropping |
| infiniflow/ragflow | #16419 | Jul 1 | Fix GoogleScholar agent tool returning empty JSON and ignoring top_n |
| infiniflow/ragflow | #16415 | Jul 3 | Port Crawler to ToolBase so it can load and run as an Agent tool |
| infiniflow/ragflow | #16417 | Jul 3 | Port AkShare to ToolBase so it works as an Agent tool |
| deepset-ai/haystack | #11755 | Jun 25 | Handle unknown tool names in human_in_the_loop confirmation strategies |
| deepset-ai/haystack | #11795 | Jun 29 | Raise FilterError on unknown metadata filter operators instead of silently passing |
| arc53/DocsGPT | #2567 | Jun 25 | Migrate AnthropicLLM from deprecated completions API to messages API |
| github/awesome-copilot | #1939 | Jun 12 | Add Pinecone RAG skill for pipelines and agent memory |
| emmabostian/developer-portfolios | #3478 | Apr 8 | Add Muhammad Furqan to the developer portfolios list |
I'm always open to interesting conversations — whether you're building something with LLMs, exploring Voice AI, or want to talk shop about RAG systems and AI agents.
- 💼 Roles — Open to AI/ML engineering roles at companies working on meaningful problems
- 🎓 Research — Interested in collaborating on NLP, RAG, or conversational AI research
- 📬 Get in touch — muhammadfurqan0100@gmail.com or connect on LinkedIn
Building AI systems that work in production · RAG · Voice AI · LLM Applications