Muneeb Ali Anwar
Full-Stack AI Engineer building
|
5 years shipping intelligent systems that solve real problems — from enterprise HCM AI engines and multi-agent pipelines to voice cloning and autonomous web agents.
About
Engineer who ships intelligence
I design, train, and ship machine-learning and deep-learning systems end to end — from CNNs, LSTMs and Transformers (BERT, GPT, T5) to production agentic AI with LangChain, LangGraph and MCP. I build RAG pipelines over vector databases, orchestrate multi-LLM engines with automatic fallback, and back it all with scalable FastAPI / Django / NestJS services, Docker, and CI/CD.
Five years in, currently at IKONIC Dev, engineering 17+ production AI features for an enterprise HCM platform — obsessed with agentic AI, multi-LLM fallback, and clean backends.
Years Experience
Live Products
AI Features Shipped
Public Repos
Experience
2+ years shipping AI in production
The timeline fills as you scroll.
IKONIC Dev
Jan 2026 — PresentAI Engineer
- Engineering production AI features for an enterprise HCM platform — a multi-provider LLM engine (Gemini → Groq → OpenAI) with automatic fallback for high availability and cost control.
- Building RAG pipelines with vector databases powering policy Q&A assistants, AI helpdesk, recruitment scoring, and performance analytics.
- Developing scalable backends with FastAPI and Django, containerized with Docker and shipped through CI/CD pipelines.
- Designing multi-tenant, multi-country architecture with real-time biometric attendance and productivity tracking.
Python Technologies
Oct 2025 — Jan 2026Senior AI Engineer
- Led development of AI/ML solutions with Python, TensorFlow, PyTorch; built agentic AI systems with LangChain and LangGraph.
- Designed LLM-powered apps with RAG architecture, Pinecone / ChromaDB vector stores, prompt engineering, and Transformer fine-tuning (BERT, GPT).
- Shipped production APIs with FastAPI, Flask, Docker; MLOps pipelines plus automation via n8n, Zapier, and Make.com.
- Built NLP solutions for document processing, chatbots, and intelligent automation.
GrowUp-TechSols — NASTP
Dec 2024 — Sep 2025AI Engineer
- Developed computer-vision models with Python, PyTorch, OpenCV — real-time hand-gesture detection with MediaPipe and facial emotion recognition on FER2013.
- Deployed ML/DL models via Flask REST APIs with full data preprocessing, EDA, and model evaluation.
NexoraSols
Feb 2024 — Dec 2024Machine Learning Engineer
- Built and deployed ML/DL models using Python, PyTorch, and OpenCV with Flask REST APIs.
- Designed data preprocessing pipelines — feature extraction, cleaning, and augmentation.
Projects
Production systems, live deployments
10 of these are live right now — click through and try them.
WorkPulse HCM
AI-Powered Human Capital Management Platform
17+
AI features
3
LLM providers
Multi
Countries
- Architected a multi-provider AI engine with automatic fallback powering 17+ AI capabilities across the employee lifecycle.
- Built “Buddy” — a RAG HR assistant with vector search over policies, plus an AI Helpdesk with auto-categorization and sentiment-based priority.
- Delivered AI Recruitment (LLM candidate scoring), AI Performance (bias detection, flight-risk), and an AI Learning Engine (auto quizzes, skill-gap analysis).
- Integrated shift-aware biometric attendance with real-time ZKTeco sync, multi-company / multi-country support, and productivity tracking.
Site Generator
Autonomous Agent Pipeline That Researches & Builds Client Websites
69%
Cache hit rate
$1.39
Cost / site
25+
Sites shipped
- An operator console that turns a business name into a researched, designed, deployed website — end to end, with no human writing code.
- Agents research the business, generate a content brief, build the site, push a repo, and deploy to Vercel; a human review gate holds every link before it goes out.
- Brief caching cuts repeat research spend — 69% cache-hit rate on 13 tracked jobs, driving cost down to ~$1.39 per generated site.
- Full job telemetry: per-job cost, stage tracking, status filtering, and one-click preview + repo links for every generated build.
Hakem.ai
AI Insurance Comparison Platform — Saudi Market
Seconds
Compare time
🇸🇦 KSA
Market
- Intelligent insurance comparison with PDF processing, automated quote extraction, and AI policy analysis using RAG.
- Pinecone vector search for semantic retrieval and context-aware recommendations with LangChain + GPT-4.
- Side-by-side provider scoring with premium, rate, strengths and a recommended plan, exportable as a PDF report.
- FastAPI + MongoDB backend with PDF parsing (PyPDF2, pdfplumber), deployed on DigitalOcean with Docker and CI/CD.
HouseholdProfits
Velocity AI — Multi-Agent Brain for Household Financial Governance
12+
Endpoints
householdprofits.com
Domain
- The AI brain behind a live household financial-governance product — assigns income, protects savings, and enforces spending rules before money moves.
- Multi-agent FastAPI service: a conversational onboarding agent (state-machine intake) plus a Co-Pilot chat agent with router → tools → framing architecture.
- Tiered LLM routing — a premium model for strategy and financial reveals, a fast model for routing, extraction, and summarization.
- Every LLM output is structured and Pydantic-validated; deterministic math, rules, risk-scoring, and forecasting engines kept fully separate from the LLM layer.
- Stateless, provider-agnostic design with optional Chroma vector memory and 12+ endpoints: transaction tagging, governance evaluation, couples synergy, forecasting, risk, milestones.
Voice Cloning Pipeline
Oral-History Voice Preservation — 9-Stage Automated Pipeline
9
Pipeline stages
3–15s
Clip length
22kHz
Sample rate
- End-to-end voice cloning from raw two-voice interview audio: a fully automated 9-stage pipeline from ingest to a fine-tuned per-person voice model.
- WhisperX transcription + alignment + speaker diarization, then LLM-based narrator detection (Llama 3.1 via Ollama) to isolate the right speaker automatically.
- Silence-based chunking to 3–15s clips at 22050 Hz, per-chunk Whisper transcription with text cleaning, and LJSpeech dataset building with eval splits.
- Per-speaker Coqui XTTS v2 fine-tuning driven by one idempotent CLI orchestrator with resumable status tracking — deployed to an Azure GPU VM via a self-hosted CI runner.
Verve AI
Real-Time AI Interview Copilot — Desktop & Browser
Real-time
Latency
4
Platforms
- Live interview assistant that listens to questions and streams drafted answers fast enough to use mid-conversation.
- Smart prompt routing keeps time-to-first-token low, with answers framed from the candidate's own résumé context.
- Ships across browser, macOS, Windows and mobile from one codebase, with a marketing site, pricing, and docs.
Osprix
Aviation Parts & Supplies E-Commerce — Live on Custom Domain
6+
Categories
osprix.com
Domain
- Full aviation-supply storefront — hangar supplies, engine parts, avionics, sealants, lubricants and adhesives — serving the aviation maintenance community.
- Catalog search with category faceting, brand browsing, product detail pages, cart and support flows.
- Shipped on a real custom domain with live support chat and a staged pre-launch ordering gate.
G.O.A.T CRM
Sales CRM SaaS for an Elevator & Escalator Company
6
Pipeline stages
Web + iOS + Android
Platforms
- Full sales-pipeline CRM — leads → site visit → quote → negotiation → won/lost — with dashboard analytics, reminders, and quick status changes.
- Interactive 3D floating elevator scene on the login page built with react-three-fiber.
- Domain engineering calculators: hoistway dimensions → rated capacity mapping from manufacturer tables, plus a price calculator.
- JWT-auth FastAPI backend, Android & iOS builds from the same codebase via Capacitor, deployed on Vercel + VPS with a server-side API proxy.
Addy — Advisor Console
Multi-Spoke AI Advisor with Human-in-the-Loop Review
8
Agent spokes
Never
Auto-send
- Internal advisor console: pick a client and Addy loads their brief, meeting notes and pipeline straight from Notion — no copy-paste.
- Eight specialist spokes from marketing to document automation, coordinated by one advisor persona that drafts in a consistent voice.
- Human-in-the-loop by design — Addy only drafts; every output is reviewed and released by an operator, nothing sends automatically.
gimme
Scroll-Driven 3D Food Ordering Experience
Scroll-to-spin
Interaction
WebGL 3D
Render
- A food-ordering site built as an experience — scroll drives a real-time 3D product spin instead of a static hero image.
- Narrative scroll journey from craving → kitchen → on the way → delivered, with live order tracking framing.
- Heavy WebGL work kept smooth with staged loading and a deliberate “enter the experience” gate.
AdMotion
DOOH Advertising Platform — Final Year Project
AI route-aware
Targeting
Live GPS
Tracking
- Out-of-home advertising platform for car roof-mounted screens — advertisers create location-aware campaigns.
- An AI engine decides which ad runs where, when, and on which vehicle, and generates campaign insights.
- Separate advertiser and driver portals plus a full admin dashboard for vehicles, ads, campaigns, and analytics.
Activity
I ship every week
Live from my GitHub contribution graph — not a screenshot. Hover any square for that day.
Skills
The stack
Click any module to expand it.
Let's build something intelligent
Open to AI engineering roles, agentic systems work, and ambitious freelance builds. Pick a channel — I reply fast.