# ced.ai — AI Consulting Firm > ced.ai is an AI consulting firm that builds production-grade AI systems for businesses. Founded by Cedric Williams (UNC Chapel Hill CS/Econ, Georgia Tech MS HCI, Data Scientist since 2012). ## Core Identity **Equation:** ced.ai = CompSci + CogSci + AI Three disciplines, one methodology. Computer science for engineering rigor. Cognitive science for human-centered design. Artificial intelligence applied strategically where it reduces friction. **Founder:** Cedric Williams - BS Economics + Computer Science (AI/ML Concentration), University of North Carolina at Chapel Hill - MS Human-Computer Interaction, Georgia Institute of Technology - Data Scientist since 2012 - LinkedIn: https://linkedin.com/in/cedricwilliams - GitHub: https://github.com/cedricwilliams --- ## 6-Step Methodology Every engagement follows this framework: 1. **HCI / Wireframing** (Origin: Georgia Tech MS HCI) — Understand users before writing code. Observe actual workflows, interview stakeholders, map the real process. 2. **SpecKit / Specify** (Origin: Software Architecture) — Define every input, output, edge case, and failure mode before building. 3. **Pure Functions** (Origin: Computer Science) — Modular, testable, debuggable components. No side effects, no hidden state. 4. **Test-Driven Development** (Origin: Engineering Discipline) — Write tests before code. 406 tests on bid automation, 200+ on engagement tracker. 5. **Regression Testing** (Origin: Production Experience) — Catch drift before users do. AI systems degrade silently without it. 6. **CI/CD with AI Agents** (Origin: Modern AI Ops) — Automated quality gates, evaluation pipelines, continuous improvement. --- ## Three Pillars ### CompSci — Systems That Hold Up Engineering rigor: pure functions, TDD, regression testing, CI/CD. The software backbone that makes AI systems debuggable, testable, and production-ready. A chain of pure functions means each stage is independently testable and replaceable. ### CogSci — Design for Humans Human-centered design: trust surfaces, workflows, correction loops, adoption strategy. A trust surface is where a human decides to accept or override the AI's output. Every AI output should be traceable, overridable, and show its confidence level. ### AI — Leverage Where It Matters Strategic AI application: extraction, classification, summarization with traceability, automation that knows when to refuse. LLMs for bounded tasks. ML pipelines for pattern recognition. AI agents for orchestration. A regex that works is better than an LLM that usually works. --- ## Case Studies ### SSA / Bid Manager (Construction x AI) - **Problem:** 4-hour manual bid process, ~2 bids/day, scattered across email/Drive/Sheets - **Solution:** AI email classification (Claude 3.5 Sonnet) + document analysis + knowledge graph - **Results:** 90%+ time savings, >95% classification accuracy, 406 tests - **Tech:** n8n, Python, FastAPI, React, Claude AI, OpenRouter, Google Workspace APIs, Firestore - **Methodology:** All 6 steps applied - **Key insight:** The documented workflow had 6 steps. The real workflow had 23. - **URL:** https://ced.ai/work/bid-manager ### NCCA Zoom Tracker (Education x Analytics) - **Problem:** Virtual school losing 25% of class time to camera policing. Zoom has no camera on/off API. - **Solution:** Python library inferring camera status from QoS bitrate metrics (3-tier detection) - **Results:** 25% instructional time reclaimed, 200+ tests, zero false positives, deterrent effect - **Tech:** Python, Flask, Zoom API, SSE, Server-to-Server OAuth - **Methodology:** Steps 1, 3, 4, 5 - **Key insight:** The problem wasn't "we need camera tracking." It was "I lose my first 15 minutes every class." - **URL:** https://ced.ai/work/ncca-zoom ### Merces Watch App (Consumer x Mobile) - **Problem:** No intelligent watch face discovery for Apple Watch - **Solution:** AI camera color extraction + color theory matching + custom .mercesface format - **Results:** 71 Swift files, MVVM architecture, offline-first, pre-launch - **Tech:** Swift, SwiftUI, Firebase, AdMob, WatchConnectivity, ZIPFoundation - **Methodology:** Steps 1, 2, 3 - **Key insight:** People choose watch faces by context ("goes with this outfit"), not by category. - **URL:** https://ced.ai/work/merces --- ## Insights Thought leadership on disciplined AI engineering — pure functions, TDD, and regression testing for production-grade agentic AI systems. ### Pure Functions: The Missing Discipline in Agentic AI - **Topic:** Why pure functions make AI agent pipelines testable, debuggable, and replaceable - **Key argument:** Most agentic AI systems are untestable black boxes. Pure functions fix this — each agent step becomes independently testable and replaceable. - **Proof point:** 406 tests on a production AI system caught 3 classification edge cases during a model upgrade. - **URL:** https://ced.ai/insights/pure-functions-agentic-ai --- ## Services - **AI Strategy** — Assessment and roadmapping for AI adoption - **System Development** — Full-stack AI system engineering - **Workflow Automation** — Automating manual processes with AI - **Technical Assessment** — Evaluating existing systems and AI readiness --- ## Key Concepts **Trust Surface:** The point where a human decides to accept or override AI output. Designed explicitly with transparency, override capability, feedback loops, and confidence signals. **Pure Function Pipeline:** AI architecture where each processing stage has no side effects. Input in, output out. Makes AI systems debuggable when things go wrong. **Monday Morning Test:** Design principle that systems must work for tired, rushed, skeptical users — not just demo audiences. If it requires training, it's not simple enough. **AI with Traceability:** Every AI output traces to source data and reasoning. Users see what the AI did and why. Essential for trust and meaningful human oversight. --- ## Frequently Asked Questions **Q: What does ced.ai do?** A: ced.ai is an AI consulting firm that builds production-grade AI systems for businesses. We combine computer science, cognitive science, and AI to automate workflows and solve real business problems. Clients typically see 90%+ time savings on manual processes. **Q: What is the ced.ai methodology?** A: A 6-step framework: (1) HCI/Wireframing, (2) SpecKit specification, (3) Pure Functions, (4) TDD, (5) Regression Testing, (6) CI/CD with AI Agents. Ensures systems work in production, not just in demos. **Q: What industries does ced.ai work with?** A: Construction (bid automation), education (analytics platforms), consumer technology (AI mobile apps). The methodology applies across industries. **Q: How does ced.ai use LLMs?** A: For specific, bounded tasks: email classification, document extraction, summarization with traceability. Each LLM call has defined input, expected output, and evaluation criteria. >95% classification accuracy on production systems. **Q: When should you NOT use AI?** A: When straightforward automation solves the problem, when error tolerance is zero, when data quality is too poor, or when adding AI creates more complexity than it removes. **Q: What is a trust surface?** A: The point where a human decides to accept or override AI output. ced.ai designs these explicitly — transparent, traceable, overridable. **Q: What credentials does ced.ai have?** A: Cedric Williams holds Economics + CS (AI/ML) from UNC Chapel Hill and MS HCI from Georgia Tech. Data Scientist since 2012. **Q: How do I start working with ced.ai?** A: Contact form at ced.ai/contact or email hello@ced.ai. We reply within 48 hours with an honest assessment of fit. --- ## Contact - **Website:** https://ced.ai - **Email:** hello@ced.ai - **LinkedIn:** https://linkedin.com/in/cedricwilliams - **GitHub:** https://github.com/cedricwilliams - **Contact Form:** https://ced.ai/contact --- ## All Pages - Home: https://ced.ai/ - Approach (Methodology): https://ced.ai/approach/ - CompSci Pillar: https://ced.ai/approach/compsci/ - CogSci Pillar: https://ced.ai/approach/cogsci/ - AI Pillar: https://ced.ai/approach/ai/ - Insights: https://ced.ai/insights/ - Pure Functions in Agentic AI: https://ced.ai/insights/pure-functions-agentic-ai/ - Case Studies: https://ced.ai/work/ - SSA / Bid Manager: https://ced.ai/work/bid-manager/ - NCCA Zoom Tracker: https://ced.ai/work/ncca-zoom/ - Merces Watch App: https://ced.ai/work/merces/ - FAQ: https://ced.ai/faq/ - Contact: https://ced.ai/contact/