Edris Paikan

CTO & Co-Founder @ BRAIKE — Levallois-Perret / Paris, France

AI & Data Engineer turned founder, based in Paris. I co-founded BRAIKE, an AI-augmented paid media agency, where I lead the technology: agentic products that give advertisers back control over their data, their tools and their decisions.

What I Do

GEO: Generative Engine Optimization

Making a brand visible inside AI-generated answers. Since AI Overviews, ranking first is no longer enough: what counts is being quoted when the machine answers in the engine's place.

AI Search & brand visibility

Measuring and steering how AI assistants describe a brand: mentions, sentiment, competitive positioning and which sources they actually cite.

Agentic systems & MCP

Designing agents that drive real business tools, with the guardrails that make them defensible in production.

RAG & knowledge systems

Building assistants that answer from your documents, with sourced and traceable answers rather than confident guesses.

Paid media & marketing data

Consolidating, measuring and automating media buying, with data that belongs to the advertiser and decisions you can audit.

AI governance & compliance

Making AI systems explainable and compliant, treating traceability as an architecture constraint rather than a legal notice bolted on afterwards.

What I Build

The BRAIKE Suite, AI Products for Paid Media

Pilot, Conversational Ad Operations Copilot

An AI copilot that reads, analyzes and edits Google Ads, Meta and TikTok campaigns live, by chat or voice. Built on an agentic architecture with 17 active connectors, ad platforms, GA4, Search Console, CRMs and file storage. Every change requires explicit human approval, and the full audit log is transparent.

Why it matters: Advertisers pilot their campaigns in natural language, without ever handing over control of the account.

CheckUp, Continuous Ad Account Auditing

Automated auditing engine that continuously inspects advertising accounts and competitive positioning. Surfaces structural flaws, misconfigurations and missed opportunities that manual reviews consistently miss.

Why it matters: Turns account audits from a quarterly slide deck into a continuous, verifiable process.

Sentinel, Media Budget Watchdog

Read-only monitoring system watching paid media spend 24/7 across Google Ads, Meta, TikTok, Amazon, LinkedIn and Microsoft Ads. Computes hourly spend projections, detects broken pixels and silent tracking failures, and pushes AI-explained alerts to Slack, WhatsApp, Teams, SMS or email.

Why it matters: Catches runaway budgets and conversion drops in real time, before they turn into invoices.

Multi-Model AI Infrastructure

The model layer powering the BRAIKE suite, routing across Anthropic, OpenAI, Gemini, Mistral and DeepSeek depending on the task, with European hosting, encrypted data and OAuth-based access that never exposes client credentials.

Why it matters: An auditable, European alternative to the black boxes of media buying.

Founded Products

Findjob, Apprenticeship Search, Automated

A platform that automates the apprenticeship hunt for students, letting them send applications at scale against a database of over a million companies. Founded and led as CEO since February 2026.

Why it matters: 100+ students placed, 1M+ companies indexed.

VoicIA, Generative AI Visibility Platform

Founded at HEROIKS: a Generative AI visibility and intelligence platform tracking how brands surface inside AI-generated answers. Built around GEO (Generative Engine Optimization) and AI search monitoring.

Why it matters: Measures brand presence in AI search, where classic SEO metrics go blind.

Peakace.app, SEO Tooling Suite

Built the Peak Ace tooling platform: an SEO suite combining automated auditing, SERP analysis and AI-assisted content workflows for large-scale media operations.

Why it matters: Automates the audit and production work that scales badly by hand.

AI & Data Engineering

MCP Connectors Hub

A Model Context Protocol meta-server connecting Google Ads, Meta, TikTok, Notion, HubSpot, Slack and more, so AI agents can operate real business tools through one standard interface.

Why it matters: Lets organizations plug AI into existing tools without building every integration in-house.

RAG Knowledge Systems

Retrieval-augmented assistants built over internal documentation, FAQs and databases, delivering sourced, up-to-date answers instead of confident guesses. Deployed for support, internal documentation and knowledge management.

Why it matters: Cuts internal information retrieval time and keeps answers traceable to their source.

Data Pipelines & Automated BI

Airflow and BigQuery ETL pipelines feeding Looker and Power BI dashboards, consolidating scattered enterprise data into a single automated source of truth. Built at scale during the HEROIKS and Didaxis years.

Why it matters: One automated source of truth instead of a dozen conflicting exports.

Text-to-SQL BI Assistant

Interactive dashboards where decision makers ask questions in plain language; the system resolves them against the data warehouse and generates contextual answers via LLM.

Why it matters: Democratizes data access for people who do not write SQL.

EOLYS Platform

Full-stack platform built for Didaxis, covering the operational needs of an umbrella-company business, from relational data modeling to the front-end used daily by consultants.

Why it matters: Core operational platform, built and maintained over two years.

AI API Marketplace (RapidAPI)

Designed and monetized custom AI endpoints, sentiment analysis, classification, embeddings, summarization, packaged as commercial APIs on RapidAPI.

Why it matters: Direct monetization of AI capabilities as productized APIs.

Generative AI for Content Production

AI Video & Image Generation Pipeline

End-to-end pipeline for automated creative asset generation using Google Veo 2, Imagen 3, and Runway Gen-3 for fashion and luxury brands. Automated product shoots, video ads, and campaign visuals at scale.

Why it matters: Reduced creative production costs by 70% while tripling output volume.

Brand Asset Management with GenAI

Intelligent asset management platform powered by AI for automated tagging, cataloging, and creative generation. Manages 50K+ assets with AI-powered search and automatic variant generation for multi-channel campaigns.

Why it matters: 60% faster campaign deployment across 30+ markets.

Dynamic Ad Creative Automation

Real-time generation of personalized ad creatives using GenAI models. Auto-generates thousands of ad variants (images, copy, video) tailored to audience segments and A/B tested at scale.

Why it matters: 3.2x improvement in ROAS through hyper-personalized creative.

AI Search & Semantic Intelligence

Semantic Product Search Engine

Vector-powered intelligent search replacing traditional keyword search. Understands natural language queries, product attributes, and user intent to deliver hyper-relevant results. Deployed across e-commerce catalog of 2M+ SKUs.

Why it matters: +40% search conversion rate, -60% zero-result searches.

Enterprise Knowledge Graph & AI Search

Built a company-wide knowledge graph connecting internal docs, Confluence, Slack, and databases. Employees ask questions in natural language and get precise answers with source citations.

Why it matters: Reduced internal information retrieval time by 75%.

AI Transformation Programs

AI Maturity Assessment & Roadmap

Comprehensive AI transformation consulting for enterprise value chain optimization. Audited 12 business units, identified 40+ AI use cases, and delivered prioritized implementation roadmap with ROI projections.

Why it matters: €15M projected annual savings across identified AI use cases.

Supply Chain AI Optimization

ML-powered demand forecasting and inventory optimization system. Reduced stockouts by 35% and overstock by 28% through real-time demand prediction integrating weather, events, and market signals.

Why it matters: €8M annual savings in inventory optimization.

Customer 360 & Predictive Analytics Platform

Unified customer data platform with predictive models for churn, lifetime value, and next-best-action. Consolidated data from 15+ sources into a single customer view powering personalized experiences.

Why it matters: 22% reduction in churn rate, +18% customer lifetime value.

GEO & Sentiment Analytics

GEO Analytics, Voice & Sentiment Engine

Advanced voice and sentiment analysis platform processing customer calls, social media, and reviews. Extracts emotions, intent, satisfaction scores, and actionable insights in real-time across 8 languages.

Why it matters: Real-time customer sentiment tracking across 500K+ monthly reviews.

Generative Engine Optimization (GEO) Suite

Proprietary toolkit for optimizing content visibility on AI-powered search engines (ChatGPT, Perplexity, Google AI Overviews). Analyzes how LLMs cite and rank content, then optimizes for maximum AI visibility.

Why it matters: +180% visibility on AI-generated answers for top clients.

LLM Visibility & AI SEO

LLM Visibility Tracker & Optimizer

SaaS platform monitoring how brands appear in LLM-generated responses (ChatGPT, Claude, Gemini, Perplexity). Tracks brand mentions, sentiment, and competitive positioning across all major AI assistants.

Why it matters: First-to-market tool adopted by 200+ digital agencies.

AI Content Authority Builder

Platform that analyzes, structures, and enriches web content to maximize citation probability by AI models. Uses reverse-engineering of LLM training data patterns and citation behavior to optimize content authority.

Why it matters: 3x increase in AI citation rate for optimized content.

Track Record

CTO & Co-Founder — BRAIKE (June 2026 - Present)

AI-augmented paid media agency. I lead technology and build the product suite.

CEO & Founder — Findjob (February 2026 - Present)

Platform automating apprenticeship applications at scale for students.

AI & Data Engineer, Founder of VoicIA & Peakace.app — HEROIKS (October 2024 - February 2026)

AI and data engineering across the Heroiks Group and its subsidiaries, including VERSUS Agency.

Data Consultant — Independent (September 2022 - January 2025)

Independent data and software consulting.

Data Engineer — Didaxis (August 2023 - September 2024)

Data engineering and process automation.

Software Engineer — Didaxis (October 2021 - August 2023)

Full-stack development on the EOLYS platform.

Certifications & Education

Bachelor's in Computer Science (Licence informatique générale L3) — Conservatoire National des Arts et Métiers (CNAM) (July 2023 - July 2024). Graduated with Excellent standing. Data analysis and Business Intelligence.

Bachelor's, Digital Innovation & IT, AI and Big Data track (E3IN) — ESIEE-IT (August 2023 - July 2024). Big Data, data analysis, artificial intelligence.

PL-300: Microsoft Power BI Data Analyst — Microsoft (February 2023). Data modeling, DAX, report design and analytics on Power BI.

Microsoft Certified: Power Platform Fundamentals — Microsoft (January 2024). Power Platform fundamentals, Business Intelligence and Big Data.

Notes & Takes

AI Overviews in France: the click is no longer the metric

Google has been showing AI summaries in France since 22 July 2026. Across 300,000 queries, Ahrefs measured an average 34.5% drop in click-through to the first organic result, and 26% of sessions now end on the results page when a summary appears.

My take: We spent fifteen years optimizing for a position. The position still exists, it just no longer pays the same traffic. My advice to advertisers: stop steering SEO by ranking, steer it by citation. The question is no longer "am I first", it's "does the machine quote me when it answers in my place". And if your informational traffic collapses while your product pages hold, that isn't a penalty, it's the market telling you where your value actually sits.

Source: Elorion : Impact of AI Overviews on SEO

GEO: factual density is the new internal linking

LLMs preferentially cite content carrying specific, verifiable, sourced claims. A page stating "our 2025 benchmark across 412 enterprise deployments measured a 38% reduction in time-to-resolution" gives a generative engine exactly what it needs to quote.

My take: This is the best SEO news in a decade and nobody treats it that way. For years, hollow well-optimized content beat dense badly-optimized content. Generative engines invert that: they need quotable substance. A dated figure, a method, a sample size. Practically: if you can't pull three sentences from your page that stand on their own out of context, the AI won't cite you. Write to be excerpted.

Source: Omnibound : GEO Statistics 2026

LLM traffic converts better, and that makes sense

Visitors arriving from LLMs reportedly convert at 15.9% from ChatGPT, 10.5% from Perplexity and 5% from Claude, against 1.76% for classic organic search.

My take: Careful with how you read those numbers. It isn't that AI sells better: it's that AI filters upstream. Once the assistant has answered the scoping questions, whoever clicks is at the end of the journey, not the start. You aren't gaining a better channel, you're losing the entire top of funnel and keeping the bottom. Excellent for conversion rate, brutal for volume. Teams celebrating the first without watching the second are in for a bad year-end.

Source: Omnibound : GEO Statistics 2026

The real GEO number: 92% intend, 40% execute

92% of marketers say they plan to optimize for AI search, but only 40.6% actually do it today.

My take: That fifty-point gap is the window. It won't stay open: the same thing happened with mobile in 2013 and video in 2017, and in both cases those who moved during the gap took a lead nobody caught. The cost of entering GEO today is trivial compared to what it will be once everyone structures content for citation. If you're arbitrating one budget this quarter, put it here.

Source: Omnibound : GEO Statistics 2026

Governed autonomy: the only model that holds in media buying

Per 2026 industry analysis, the gap between vendor claims and independent verification remains wide. The teams winning build governed autonomy: spend caps, approval gates and audit trails around an AI that is capable but not yet trustworthy alone.

My take: That's precisely the bet we made at Braike, and I'll state it plainly: the agent owns tactics, not strategy. A system that edits an ad account without human validation isn't progress, it's risk transferred onto the advertiser. That's why Pilot never touches an account without an explicit green light, and why Sentinel is read-only by construction. That's not commercial caution, it's the only architecture still defensible the day something goes wrong.

Source: TensorOps : Agentic AI in Advertising, 2026 Field Guide

150 decisions per campaign per day: who reviews them?

Meta Advantage+ campaigns steer over $12 billion in annual spend, with AI making 150+ optimization decisions per campaign per day.

My take: Nobody reviews 150 daily decisions. That's the blind spot in the current debate: we argue about the quality of the machine's decisions when the real problem is that they've become invisible. An advertiser who can't answer "why did my budget go there" hasn't delegated media buying, they've lost sight of it. The challenge of the next two years isn't an AI that decides better, it's an AI that accounts for itself.

Source: eMarketer : FAQ on AI media buying

Every ad platform shipped an MCP server. Now what?

TikTok launched its Ads MCP server in May 2026, letting third-party agents plan and optimize campaigns, after Google, Meta and Amazon had already shipped equivalent protocols.

My take: Ad platforms opening their own door to third-party agents is the most underrated signal of the year. They aren't doing it out of generosity: the orchestration layer is slipping away from them and they'd rather write its standard. For an advertiser it changes everything, multi-platform stops being an integration project and becomes a tooling choice. That's exactly why we built a connector hub rather than one integration per platform.

Source: TensorOps : Agentic AI in Advertising, 2026 Field Guide

MCP won: 41% of organizations in production

41% of surveyed software organizations run MCP servers in limited or broad production. Gartner projects 75% of API gateway vendors will ship MCP features by end of 2026, and Anthropic, OpenAI, Google and Microsoft have all integrated it natively.

My take: When the four players who fight over everything adopt the same protocol, the debate is over. What I tell the engineering teams I meet: stop evaluating MCP, start implementing it. The cost of not doing so isn't technical, it's strategic: every proprietary integration you write today is debt you'll repay in 2027. The question is no longer "which protocol" but "which tools do I expose, and behind what guardrails".

Source: Digital Applied : MCP Adoption Statistics 2026

MCP goes stateless: the moment it got serious

The 28 July 2026 MCP specification moves the protocol to a stateless architecture, making agent infrastructure cacheable, routable and scalable like the rest of the web.

My take: It's the most structural change of the year and it went almost unnoticed because it isn't spectacular. Statelessness is what let the web scale planet-wide; applying it to agents is admitting we're leaving prototype territory for production. Concretely, agent architectures can now sit behind standard CDNs and load balancers. Anyone who built on persistent sessions is going to rewrite.

Source: Model Context Protocol : 2026-07-28 Specification

The AI Act's Article 50 is live, and almost nobody is ready

Since 2 August 2026 the AI Act is fully applicable. Article 50 requires chatbots and conversational agents to clearly signal to users that they are talking to an AI, and AI-generated or heavily modified content to be identifiable.

My take: I've watched a lot of teams file the AI Act under "legal, handle later". Wrong department: transparency and traceability are architecture constraints, not legal notices. You don't bolt an audit log onto a system that was never designed to produce one. Teams who built human validation and full history into their agents from day one are discovering they're compliant for free. Everyone else is rewriting.

Source: IT for Business : AI Act: what changes on 2 August 2026

High-risk systems: the December grace period is a trap

Systems already on the market before 2 August 2026 get a transition period until 2 December 2026 to comply on governance, data quality, traceability, human oversight and documentation.

My take: Four months to rebuild traceability into a system that never produced any isn't a deadline, it's the illusion of one. I advise the opposite of waiting: use the deadline as a free audit of your decision chain. Nine times out of ten, the mapping exercise reveals the company can't say who approved what, on which data. That finding is worth far more than the compliance itself.

Source: French Ministry of Economy : EU AI Regulation

Auditability will become a sales argument before it becomes an obligation

Compliance for a high-risk system requires governance, data quality, traceability, human oversight and documentation, precisely the properties a black box cannot supply.

My take: My conviction, still a minority one: within eighteen months advertisers will pick their provider on the ability to explain decisions, not just on claimed performance. Because performance claimed by the platform selling the inventory isn't evidence, it's a pitch. Auditability can actually be verified. We built our entire suite on that assumption, if I'm wrong, we'll simply have made needlessly transparent tools.

Source: Pôle d'excellence cyber, AI Act obligations, August 2026

Sovereign cloud: €180M won't be enough, and that's fine

In April 2026 the European Commission selected four European providers, including OVHcloud and Scaleway, for a €180 million sovereign cloud framework contract.

My take: Against the tens of billions hyperscalers invest annually, €180M doesn't shift the balance. But I think we're fighting the wrong battle by reasoning in capacity: the sovereignty that matters to a European company isn't owning its own AWS, it's being able to switch providers without rewriting the product. Real sovereignty is an architectural property: portability, open standards, reversible data. Not a hosting address.

Source: Stratégies : European alternatives to the tech giants

Mistral raises $830M for a datacenter: infrastructure before model

Mistral raised $830 million for its first French datacenter, while Mistral Large still trails the best American models on performance.

My take: Investing in infrastructure rather than chasing the benchmark is the right order of priorities, and it deserves saying. A slightly weaker model you can host, audit and keep available beats an excellent model you depend on with no recourse. In our architectures we route by task: American models where the reasoning gap genuinely pays for itself, European models everywhere it doesn't show. That's neither patriotism nor naivety, it's risk management.

Source: JustAI : Sovereign AI in France: the reality behind the story

Europe picked Domyn over Mistral: the signal behind the surprise

On 19 June 2026 the Commission named the EUROPA consortium, led by Italian startup Domyn with Fraunhofer, winner of the Frontier AI Grand Challenge, to build a 400B+ parameter open source model across the EU's 24 official languages.

My take: The pick surprised people who expected the French champion. I read it differently: the Union funded multilingual open source rather than a national champion, and that matches what companies actually need. An open model you can audit and self-host creates more sovereignty than a closed European one. The provider's nationality matters less than the nature of the licence.

Source: Tech Insider : Mistral, Domyn and the Frontier AI Grand Challenge

Cloud lock-in is no longer about price, it's about AI integration

Migrating a stack from AWS to OVH has become extremely complex: AWS's AI integration has reached the point where developers simply ask the AI to configure the entire setup, something OVH doesn't yet offer.

My take: It's the most effective dependency mechanism ever built, and it costs the party installing it nothing. Lock-in no longer runs through egress fees but through comfort: when the vendor's assistant writes your infrastructure, your infrastructure becomes unreadable to you. My rule at Braike: anything generated by an AI must stay readable and reproducible by a human, or it doesn't ship. It's constraining, and it's what keeps the exit door open.

Source: JustAI : Sovereign AI in France: the reality behind the story

Part of what an agency bills is automatable. Now.

Google AI Max, Meta Advantage+ and a wave of autonomous agents now run targeting, bidding, creative and budget pacing with progressively less human input.

My take: It's uncomfortable to say when you sell services, so let's say it clearly: campaign setup, reporting and much of routine optimization are no longer worth what we billed for them. Denying that means selling time that's about to vanish. What remains, and gains value, is arbitration, business judgment, the ability to tell the platform no. We built Braike on that shift: tooling to free up advisory time, not to bill execution for longer.

Source: TensorOps : Agentic AI in Advertising, 2026 Field Guide

Building tools so clients can do without us

The traditional agency model rests on lasting client dependency. Open platform protocols and agentic tools now make in-housing media operations technically feasible.

My take: We took the opposite bet to the classic model, and I stand by it: our tools are built so the advertiser can one day do without their agency, Braike included. On paper that's absurd. In practice, a client who stays because they choose to beats a client who stays because they can't leave, they're paying for expertise, not for a lock. And expertise doesn't automate.

Source: CB News : Braike launch announcement

Augmented consultant over autonomous agent: why we chose

The 2026 industry trend runs toward agents piloting campaigns alone, while independent analyses recommend guardrails: caps, approvals, audit logs.

My take: Full automation is seductive in a demo and fragile in production, for one simple reason: an agent optimizes what you gave it to measure, never what you forgot to tell it. The competitive context, the margin constraint, next month's product launch, none of that lives in the ad account. That's why we bet on consultants backed by a cohort of agents rather than on the agent alone: the machine executes, the human arbitrates.

Source: TensorOps : Agentic AI in Advertising, 2026 Field Guide

A model creates no value on its own

Each new model generation improves reasoning and reliability on long-running tasks. The 2026 frontier models are MCP-native at Anthropic, OpenAI, Google and Microsoft.

My take: The model race is fascinating and it distracts from the only place value is created: everything around the model. The data it reaches, the tools it drives, the memory it keeps, the automations it triggers, the experience built around it. I've seen teams wire the best model on the market to nothing at all and wonder why. Swapping models takes a day; building the context around one takes a year. Invest where it's long.

Source: Model Context Protocol : 2026 Roadmap

Before deploying an agent: caps, approvals, logs

Successful 2026 advertising agent deployments share three traits: hard spend caps, explicit approval gates and complete audit logs, rolled out in phases rather than all at once.

My take: If I could give teams starting out one piece of advice: build the guardrails before the agent, never after. It's counterintuitive because guardrails make no impressive demo, and that's exactly why they get postponed. An agent without a spend cap isn't an agent in testing, it's an incident waiting. We wrote Sentinel's limits before its detection logic, the order wasn't an accident.

Source: Superscale : Automate Meta ads with AI agents, 2026 playbook

Tool Catalogue

AI Models & APIs

Agents, RAG & Orchestration

Data Engineering & Warehouses

Cloud, Infra & Deployment

Development & Languages

Advertising & Media Platforms

SEO, GEO & Market Intelligence

Automation, CRM & Productivity

MLOps, Quality & Security

Creative & Content Production