Product Manager and Product Owner with 5+ years building enterprise AI tools. My background is in UX, which means I think about how systems feel to use, not just how they work. I turn powerful platforms into products people actually adopt.
I'm Manuel, a Product Manager and Product Owner based in Berlin. I started in advertising and photography, moved into UX, and then into product. I enjoy the hard problems: products with real technical depth that still need to feel obvious to the people using them.
At Lengoo, a Berlin AI startup (USD 34M raised), I owned the product that helped enterprise teams configure, train, and improve custom translation models. When Lengoo closed in 2024, I kept working at Cognigy, the Conversational AI company, turning support and customer insights into clear product priorities.
My edge is simple: I understand what AI systems are doing under the hood, and I understand why users stop trusting them. I work in that gap. I am currently open to PM and PO roles at product led companies where AI is a real product challenge, not a buzzword.
Four plus years building enterprise AI platforms, from custom model tooling to conversational AI. I know the gap between "technically works" and "actually adopted."
I start with the problem, not a feature list. I use interviews, ticket patterns, and usage data before anything hits the roadmap.
Design is how complexity becomes usable. I use UX to make decisions clearer, workflows faster, and products easier to learn.
CSAT, latency, adoption, retention. I define the outcome first, then build what moves it.
Turned a fragmented ML configuration workflow into a guided platform. Result: +15% CSAT and +20% adoption.
Built a real time feedback pipeline that cut latency by 85% and automated P1 triage across four engineering teams via Jira.
Co built a studio trusted by Adidas, Zalando, and Blinkist. 100% on time delivery across 20+ clients and 60% repeat business.
AI-powered routing system that cut misrouted tickets by 70% and got urgent issues to the right team in under 24 hours.
Built the HALOS Console brand from zero: logo, color system, and marketing assets. Then applied it consistently across product and sales.
I am open to PM and PO roles at product led companies in Berlin and remotely, especially in AI, B2B SaaS, or complex technical tooling. If you need someone who can own a problem end to end, I would love to talk.
contact@manubecerra.comI owned the AI Model Customization initiative, a core part of Lengoo's HALOS Console. The goal was simple: replace a fragile, manual workflow with a product teams could rely on.
HALOS Console had the technical foundation. But the core workflow — how teams configured and customized their ML models — was held together with scripts, tribal knowledge, and a lot of Slack messages to Engineering. There was no clear interface, no visible model state, and no way for non-technical users to act without help.
The result was a constant support load on Engineering and a product that felt inaccessible to the people who needed it most.
I combined interviews, usage signals, and a hard look at where the workflow broke down. Then I shaped the scope so we could ship improvements quickly without painting ourselves into a corner.
I ran 20+ interviews with linguists, PMs, and customer success teams — not to validate a roadmap, but to understand where the workflow was actually breaking. Most of the friction was invisible in analytics. It only showed up in conversations.
Those interviews shaped the scope. We did not build what was easiest. We built what removed the most pain.
Before any spec was written, I mapped the end-to-end model customization workflow — from a user's first action to a deployed model. That map exposed three handoff points where context was lost and users had to go find someone in Engineering.
It became the shared reference for the whole team. Designers, engineers, and CS all worked from the same picture of the problem.
We moved from monthly to bi-weekly releases. But the cadence was not the goal — the feedback loop was. Each release had defined success metrics (adoption, task completion, support volume) so we knew within two weeks whether to push forward or adjust.
That discipline is how we hit +15% CSAT without a big-bang launch.
We turned a set of scripts and tribal knowledge into a clear product experience. The goal was not "more features". It was fewer mistakes, faster iteration, and a workflow that felt safe to use.
A single view where users could see the state of every model — no more pinging Engineering to ask what was running. Status was visible, actions were clear, and edge cases were handled in the UI instead of falling through to a support ticket.
Users needed control over when models ran and what they cost. The On-demand vs Always ON toggle was a small decision with real cost implications. We made it obvious, with immediate feedback so users understood what they were switching before they switched it.
Before this, performance data lived in Kibana — accessible to engineers, invisible to everyone else. The dashboard brought the key metrics into the product so PMs, CS, and linguists could see what was happening without filing a request.
The most telling signal: teams stopped filing tickets asking Engineering how to use their own platform. The workflow was now something people could navigate on their own.
Project ECHO fixed a painful truth: feedback reached engineers days too late. I connected structured linguist feedback to the backlog so issues showed up fast, clearly, and in the right order.
We had systemic latency in the MT feedback loop across four teams: Linguistics, Front End, Data Engineering, and ML. The manual process added days of delay between a linguist spotting an issue and an engineer acting on it. That delay directly hurt quality and iteration speed.
The friction was pushing expert linguists out of the loop, which meant the models lost high-quality training signal. Once linguists stopped flagging issues, the data degraded silently. The mandate was clear: remove the latency, automate triage, and keep the feedback loop alive.
Vision: Turn the feedback to fix cycle from a bottleneck into an advantage. Near zero latency iteration and an always on quality signal.
Three concrete deliverables:
I worked with Data Science to define a structured error taxonomy. We used Kibana and Datadog to identify the top 5 error classes driving 80% of rework. Those became the initial schema.
Result: objective, unambiguous Definition of Done. No more backlog refinement debates about severity.
I prioritized the data contract (JSON schema plus API endpoints) as the first sprint deliverable. That unlocked parallel work across all four teams and removed a major dependency risk.
I wrote the technical spec for the ingestion layer and the Jira integration as a single source of truth. That clarity reduced back and forth and made sprint planning faster.
When you remove friction from a feedback process, people use it. Linguist participation went back up, error trends became visible in real time, and Engineering stopped triaging tickets manually.
I co-founded Aneekaa and ran it for nearly a decade. Brand, web, photography, and video for clients who needed things done properly. Every project from scratch. Every project delivered.
Aneekaa was built from nothing. No clients, no reputation, no safety net. Just a clear idea of what good creative work looks like and the discipline to deliver it consistently.
We worked with restaurants, startups, cultural venues, and international brands. The brief was always different. The standard was always the same.
"Strategy without execution is theory. Execution without strategy is noise. Running Aneekaa taught me to hold both at the same time."
I owned every engagement end to end. Discovery, scope, creative direction, production, delivery, and sign-off. No handoffs without full context. No dropped balls between stages.
Managed designers, developers, photographers, and videographers across Spain and Germany. Tight deadlines, clear ownership per task, and a culture of just getting it done.
60% repeat business is not luck. It comes from accurate estimates, honest communication, and delivering what you said you would. Clients returned because they knew what to expect.
That was the rule. Every brief approached fresh. It was slower and more expensive to run that way. It was also the only thing that kept the quality consistent across nine years.
"Running a studio for nine years gave me something most people don't have: I've felt the full weight of delivery. I know what it means when a deadline slips, scope creeps, or a client loses confidence. That experience shapes everything I do now."
© 2026 Manuel Becerra · Berlin
LinkedInP1s were sitting in the queue for 72 hours. The same feature request arrived five different ways and never hit the roadmap. I mapped where the signal was dying and built the routing layer to stop it. Misroutes dropped 70%. Urgent issues now land in under a day.
On a contact-center platform, tickets arrived as raw text from Zendesk/Intercom, Slack, and email. Every one meant jumping into Kibana/Grafana to reproduce, then tailoring the handoff: Eng wanted logs + repro, PM needed a story, CS needed KB updates. Resolution times stretched, delays piled up, PM visibility stayed low.
Zendesk, Slack, email, PM DMs. No single intake and no shared format.
Engineering got UX nits, PMs got crash reports. P1s sat 72 hours before anyone saw them.
The same feature request arrived five different ways and never reached the roadmap because it looked different every time.
"The routing problem wasn't technical. It was structural. We didn't need a new tool; we needed a shared language for classification and a clear routing contract between teams."
The before state: invisible chaos. Work happened but not in order, not by the right people, not fast enough. The after state: one contract, one intake point, deterministic routing.
Four categories, three severities, one owner per combo. Published as the Signal Contract and signed by Eng, PM, CS, and Design.
Small in-house classifier checks the ticket, suggests taxonomy + severity, and adds a one-line rationale so the receiving team acts without clarifying.
Slack handoffs with rationale + confidence; P1s open Jira automatically. Everything is logged so humans can override quickly.
We skipped a new UI. Instead: Slack workflow + tiny classifier + Notion contract running on signals. Built over two sprints, adopted in sprint three, now feeds the backlog every week.
The core value of Signal is behavioral change: getting four teams to use a shared language and trust a shared routing decision. That is a change management problem first, a tooling problem second. A polished internal product would not have solved the adoption challenge any faster.
Signal v1 was the right scoping call. A full product makes sense once the taxonomy has proven stable across 3+ months, teams are asking for analytics on signal trends, and there is a business case for extending routing to external channels.
Three decisions shaped Signal. Each one is the kind of call a PM has to own. No committee, no consensus. Just a clear reason and the willingness to defend it.
Signal was never a product. It was a system of agreements backed by just enough automation to make those agreements stick.
// Manuel Becerra · PM · 2024
I built the HALOS Console brand from scratch: logo, color system, and marketing materials. Then I applied that visual language across the product, website, and sales assets so everything felt like one coherent system.
HALOS Console was Lengoo's enterprise AI platform. The tech was strong, but it had no visual identity. It needed a brand that worked in product UI, marketing, sales decks, and external comms while still feeling credible to enterprise buyers.
I owned this end to end: positioning, logo construction, color system, and applied assets. No agency. Built in house from concept to production.
The primary brand expression: the "hi" icon paired with the HALOS Console wordmark. Designed to work at any scale, from product UI to marketing.
The logo works on light and dark surfaces. Both versions use the same icon. Only the wordmark color changes so it stays legible everywhere.
The "hi" icon works as a standalone mark for small formats like product UI, favicons, app icons, and social. It is built on a geometric grid so it scales cleanly from 16px upward.
The logo went through an exploration phase with geometric construction systems, gradient directions, and letterform variations. The construction grid shows how the curves and proportions come from the same circle geometry.
Three primary brand colors, each with a role. Cyan (#0abce9) for energy, navy (#3f4497) for trust, and blue (#007fe4) as the connector. Together they work across UI, print, and digital.
Building a brand from scratch forces decisions to be intentional. Color had to work in a data table, a loading state, a marketing headline, and a small icon badge at the same time. That constraint driven thinking is exactly how I approach product design.
The same visual system became the foundation for the HALOS Console product UI: color tokens, type hierarchy, and icon style. Brand and product were not separate workstreams. They were one.
"A brand is not a logo. It is a set of decisions that makes every touchpoint feel like it came from the same place, from the favicon to the sales deck. I built that system for HALOS Console and then shipped the product on top of it."
