Lahore, Pakistan · Full-Stack AI Engineer @ Sensus DWC LLC (Dubai) · MS Artificial Intelligence student @ LUMS
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'm also studying for an MS in Artificial Intelligence at LUMS. 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
- ✅ Evals & guards: LLM-as-judge and rule-based eval suites, plus server-side checks around every agent action
Open to AI/ML roles and research collaborations.
|
Client: Proton Malaysia 
|
Podcast-native RAG platform for military intelligence content (built for Podlogix). Transcribes, embeds, and semantically searches a large library of military podcasts, using BM25 plus dense search and reranking behind a LangGraph chatbot. Built with Deepgram, Pinecone, and LLMs. |
💬 RMA Motors: WhatsApp RAG Chat AgentOutbound trade-in campaign RAG chat agent for an automotive dealership. Runs full conversation flows covering qualification, objection handling and 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. Each company's data, billing included, is kept separate through JWT-based tenant scoping. Built on Django REST + LangChain and deployed on AWS (EC2, RDS, S3).
|
📘 MathWizAdaptive EdTech platform built on an NCERT-based RAG pipeline for grade and chapter-level question generation. It 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.
|
Voice agents, chat agents and AI modules for car dealerships across Malaysia, the UAE, Saudi Arabia, Japan and Germany. Every agent ships with evals or server-side checks.
| Client | Project | What it does |
|---|---|---|
| Regal Motors | Arissa on WhatsApp | RAG chat agent on Claude Haiku 4.5 with a six-action JSON trigger layer and a 109-case eval suite (code checks plus a validated LLM judge) |
| Cycle & Carriage | Leapmotor agent | Outbound WhatsApp agent in English, Malay and Chinese that follows up leads and books test drives. The server checks every trigger and required field before a booking is logged |
| Arcfox UAE | Dealership RAG chat agent | Claude tool calls over PostgreSQL with weighted fallback search, pgvector semantic search and a 74-case eval suite |
| Service My Car | Sara | Claude Haiku agent with an LLM-as-judge and rule-based eval suite (68 cases) |
| Nissan Tokyo | Yui | Japanese outbound maintenance-booking voice agent where every call ends in a logged outcome |
| Nissan Saudi Arabia | Nissan Petromin chatbot | Service booking for 15 centres in 7 cities, with date and time validation and multi-location routing |
| Dünnes | Inventory API for a voice agent | Read-only PostgreSQL API with 28 validated search filters, bearer auth and ElevenLabs webhook tools |
| TM Automotive Group | TMG chatbot | Dealership chatbot for used-car and trade-in enquiries |
| ieSync | Workshop Manual Agent (demo) | RAG over BMW and Mercedes repair manuals (Pinecone, Gemini embeddings, Claude) that answers with the manual's own step photos. Retrieval tests: 0 failures across 33 questions |
| Titan DMS | Sales Forecasting API, Deal Profitability API | FastAPI services: one backtests 5 models per request and returns a forecast with confidence intervals, the other predicts gross profit or the chance a deal loses money |
| 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
- 🏗 Named Haystack 3.0 Community Contributor: publicly recognised by deepset in the official Haystack 3.0 launch post alongside core team members (Julian Risch, Stefano Fiorucci, and others)
- 🥇 Kaggle Notebooks Master: rank 881 of 59,179 (peak rank 149)
- 🌍 LinkedIn Top Voice (AI): Recognised 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 Science Professional
- 🛠 Building: voice and chat agents for car dealerships at Sensus, with evals and guards around every agent
- 🎓 Studying: MS in Artificial Intelligence at LUMS
- 📚 Reading: AI Engineering by Chip Huyen
- 🎯 GSoC 2027: Building contribution history toward Google Summer of Code (target: mem0ai/mem0)
- 💼 Open to: AI/ML engineering roles · Research collaborations
20+ merged PRs across major AI/ML frameworks. The tables list the main ones.
| 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 |
| Repository | PR | Merged | What was added |
|---|---|---|---|
| NousResearch/hermes-agent | #70512 | Jul 24 | Add pinecone-research optional skill for agent RAG and long-term memory; cherry-picked and merged by teknium1 |
| github/awesome-copilot | #1939 | Jun 12 | Add Pinecone RAG skill for production pipelines and agent memory |
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