I'm Ahnaf Thaqeef. I design and build AI systems, RAG pipelines, and automation workflows that cut costs and generate real business results.
I'm obsessed with one question: how can AI make businesses meaningfully better?
From voice agents that handle customer calls to tools that write content, generate resumes, and automate workflows — I build end-to-end AI products with a focus on real-world outcomes.
I work at the intersection of product thinking and technical execution. If there's a problem worth solving, I'll figure out the right AI approach for it.
A selection of AI tools and systems I've designed and shipped.
Agents, chatbots, voice, and workflow automation
AI-powered voice agent built on Retell AI and ElevenLabs — handles inbound/outbound business calls with natural conversation flow and CRM integration.
Standalone voice agent built directly in ElevenLabs — production-ready conversational AI with custom voice cloning.
AI chatbot built on Claude and a full RAG pipeline. Ingests real company data — product specs, FAQs, service info — and answers customer questions accurately. Embeds as a widget on any website. Demo: live automotive spec assistant trained on Proton X50 data.
AI-powered business proposal generator — creates tailored, professional proposals fast so you can close deals, not write documents.
Suite of AI tools for content creation — from ideation to drafting to scheduling, built for marketing teams.
Visual system for designing AI agent workflows — the WAT framework (Workflows, Agents, Tools) made tangible.
4-in-1 customer experience AI platform — agentic telco service bot, contact center analytics dashboard, GenAI ROI value case builder, and service blueprint generator. Built for SEA enterprise CX teams.
Full-stack products live and running
Upload your CV, paste a job description — get a tailored, ATS-ready resume in seconds. Built with AI-powered rewriting and Supabase auth.
Daily habit tracking with streaks, progress analytics, and smart reminders to keep you consistent.
Event management platform — attendee registration, agenda builder, and check-in workflow for end-to-end event ops.
Personal property ownership dashboard — track grant renewals, tenure expiry, tenant details, rent due dates, and portfolio value in one place.
3-in-1 finance AI app: stock sentiment analysis, economic indicator tracking, and market brief generator. Built as a portfolio piece for senior AI innovation roles.
AI-powered daily lesson plan generator for Malaysian educators. Produces structured rancangan pengajian harian in BM, aligned to KSSR/KSSM formats.
Model fine-tuning, evaluation frameworks, and cloud ML deployment
Multi-process fine-tuning of DistilBERT transformer architecture using PyTorch DDP and HuggingFace Accelerate — identical script scales from 1 CPU to N-GPU clusters. Includes DistributedSampler, all_reduce distributed evaluation, and failure mode analysis. Model published to HuggingFace Hub with full model card.
End-to-end training pipeline for a DistilBERT transformer architecture (6 layers, 66M params). LoRA fine-tuning: 592K trainable / 67.6M total (0.88%). 82.45% accuracy, F1 0.8246 on SST-2 validation. Integrated as sentiment pre-screening layer in insurance claims triage pipeline.
Novel dual-layer evaluation framework for LLM alignment — 2,000+ real inferences evaluated. Ragas automated scoring + LangSmith human-preference labelling on 312 high-confidence failures. 5-category failure taxonomy (negation blindness 27.9%, sarcasm 19.6%, code-switching 17.3%) — directly informed retraining corrections.
End-to-end ML deployment pipeline for a fine-tuned HuggingFace model: model packaging (tar.gz), S3 staging, AWS SageMaker HuggingFaceModel endpoint (ml.t2.medium), and GCP Vertex AI custom container via Docker + Google Artifact Registry. Cost-controlled teardown automation.
Documenting what works in AI for business — honest takes, no fluff.
A challenge log documenting real business problems tackled with AI tools — one per day, with what worked and what didn't.
Read More →Most businesses still use humans for calls that could be handled by AI. Here's what's actually possible today.
Read More →Workflows, Agents, Tools — a simple mental model for building AI systems that actually ship and scale.
Read More →More content coming. Follow along for daily AI breakdowns.
From clients and collaborators who've seen the work firsthand.
"Ahnaf built us a voice agent in days, not months. It handles 60% of our inbound calls now — the ROI was immediate. I didn't think it was possible this fast."
"I've worked with a lot of developers, but Ahnaf thinks like a product person. He didn't just build what we asked — he built what we actually needed. The chatbot he delivered exceeded every expectation."
"The AI automation workflow he designed saves our team 20+ hours a week. Genuinely shocked by the quality and speed of delivery. Will work with him again."
Have a business problem you think AI can solve? Or just want to talk about what's possible? I'm interested.