Forward Deployed Engineer @Tamtam · Product Ops @Harmattan AI from Feb 2027
I sit between customers and the product: I find what breaks in the field and ship the tool that fixes it.
Today I'm a Forward Deployed Engineer at Tamtam, an AI sales intelligence startup: I work inside customers' setups and turn what keeps coming up there into product. Before that I was the Growth Engineer at Finary, where I built ten internal tools that gave the teams back about 190 hours a month.
I like to understand a team's problem before I join it: three of the products below were built to get hired, and two of them did the job.
I'm in the final year of a business and engineering double degree, Audencia × Centrale Nantes (BBA Big Data, AI & Management).
Product Operations · internship
Feb 2027
Start date
Upcoming
French defense tech: drones, counter-drone and ISR.
Joining the Product & Programs team to build the tooling product managers run on: competitive intelligence, reporting and AI workflows.
Forward Deployed Engineer · freelance
3
new detectors in the product
01New opportunity signals
AI sales intelligence: agents on top of a company's CRM that find who to sell to next.
Forward deployed means working inside customers' setups: wiring their CRM, configuring the product for their case, and turning what keeps coming up into features.
Added three opportunity detectors to the product: a prospect moving away from a competitor's tool (spotted in their job posts), a new sales leader taking the job, and lookalikes of an existing customer.
Synced the customer base with the CRM and the product, then cross-checked all three and sorted every mismatch by type and confidence, so fixes are made on evidence, not gut feel.
Rebuilt the product request log out of no-code, with an agent that triages what customers ask for on behalf of customer success.
Built an agent that handles bug reports and feature requests, scoped tight for a first version so it does a few things reliably.
Growth Engineer · internship
~190 h
a month back to the teams
01Ten internal tools
Personal finance app with about 1M users in Europe.
Solo owner of Growth Data & Automation: I found where the teams lost time, built the tool, and handed it over documented, with a named owner for each system.
Ten tools built, adopted and handed over, giving the Growth and Media teams back about 190 hours a month: more than one full-time role.
Segmented a base of about 1M users and scored ~3,500 prospects, which turned into 20+ qualified meetings for the advisors. The Care team now runs it.
An expert-gated content engine: one standard built to be cited by AI assistants (GEO), a sourced knowledge base, and one article a day, with tax experts approving every new piece.
Rebuilt my application project inside Finary: checking a video script went from seven hours to ten minutes.
Rated "Exceeds expectations" at all three reviews.
Earlier
Data Analyst · internship
French online recruitment and training platform (Télégramme group).
Rotated between the Sales Analysis and Web Analysis teams, on real commercial questions: SQL, Salesforce, BigQuery, Looker Studio.
Analysed what drives the forecasts sales reps enter in Salesforce and correlated quotes with opportunity win rates, then turned the findings into recommendations for the sales team.
Checked Piano Analytics data against server logs, fixed the filters and rebuilt the Looker Studio dashboards that track conversion and visits.
I don't send cover letters. I ship the tool.
Product and competitive intelligence for a defense tech team, built in a weekend for my Product Ops interview.
Shown to my future manager, then opened by the CPO. I got the offer.
Independent project built from public sources. Not affiliated with Harmattan AI.
Python, Claude API, Next.js, Vercel
An internship search copilot for my classmates: build your candidate file once, reuse it at every step.
Success rule set before launch: if 2 testers out of 3 land an interview within six weeks, it moves to v2; 1 out of 3, I change one variable; none, I stop and find where it breaks.
Next.js, Supabase, Claude, OpenAI or Gemini (bring your own key), Vercel
Market intelligence on European neobrokers: every launch, in every market, from public data.
Python, Claude (Batches API), SQLite, Next.js, Vercel
Tailored intelligence briefs: tell Lia what you follow in one conversation, get your brief by email or Slack.
Runs my own fintech newsletter, twice a week.
Next.js, TypeScript, Supabase, Claude API, Docker, Hetzner
Two AI models argue every claim in a video script, and an arbiter settles it with live sources.
Got me the internship, then I rebuilt it in production at Finary.
Claude API, GPT-4, Web search API
Turns a week of raw prospects into the twenty accounts to call on Monday, and says why.
~466 raw prospects scored · Got the offer
Python, Claude (Sonnet), FastAPI, Next.js, Vercel
Turns every reply and engagement signal into one decision for the SDR: call now, triangulate, auto-reply or watch.
Not affiliated with Pigment.
n8n, Apollo, Hunter, OpenAI (GPT-5.5), Google Sheets, Python
Co-founder & COO
Co-founded with a friend. Version one was a marketplace matching design students with young companies; version two, a premium turnkey service run with a small freelance team. I owned prospecting, go-to-market and delivery.
Demand is not a business: both sides wanted the marketplace, but at around €50 of commission per project it couldn't pay. That's why we went premium.
Delivering was never the bottleneck, selling was. That's what pushed me to build prospecting machines, and into growth engineering.
A company run alongside studies and an internship hits a ceiling. I chose to close it properly rather than let it drift.
AI project scoping as a product: a guided interview that ends in an honest go/no-go, so non-technical teams can scope their own AI projects without hiring consultants.
A method tested by hand with colleagues at Finary, a full brand system, a live landing page and a first engine tested on a real case.
The engine worked, but the interview ran long and the AI lacked business context. Generic scoping isn't enough: it needs a vertical.
A financial cockpit for French freelancers: business (social charges, invoices, taxes) and personal wealth in one place, answering one question: after charges and taxes, how much can I invest this month?
An MVP pitched with a live demo at Centrale Nantes, a full business plan, and a public simulator in production.
Screened honestly, the real competitor was Indy, not Finary, and the effort-to-price ratio didn't hold for a solo founder yet. It grew out of Solen, my own investment tracker, where I learned that centralizing data is the easy part: advice is the value.
Building toward founding my own company.
Tools I built for myself. They're how I work.
Living project files: Claude Code hooks keep a readable status for each project and bring it back at the start of every new session.
My second brain: a Postgres and pgvector knowledge base of everything I work on, synced every day, that my agents search first.
My Paris apartment hunt on autopilot: listings scored on 24 weighted criteria, a phone alert above 70, scam checks built in.
A macOS menu bar app that logs freelance hours and keeps two Macs in sync through Google Calendar.
A daily security report on my 34 repos: safe fixes merge themselves, only when the build is green.
