Fig. 01 — AI Engineer · Lima, Peru

I take language models to production and they stay there.

I design the whole system: the retriever, the API that serves it, the infrastructure it runs on and the interface someone uses every day. Prototyping is easy; keeping it alive is the work.

Available for remote and freelance work
role
Head of Fullstack Development & AI Engineer · VentIA
stack
Python · FastAPI · React · Next.js · LangChain · AWS
How I workFig. 02
the businessproblemarchitecturedesignbackend, modelsand datainfrastructureas codethe interfacesomeone usessystem inproductionthis is wheremostprototypesstopshort

// AI work does not end in the notebook — it ends when someone uses it on an ordinary Tuesday

80%

of queries resolved with no human involved

Servinetwork
−60%

less manual handling after deploying LLM agents

VentIA
15s

QR rotation with geolocation validation

Qore
Need this for your company?The same work, explained without the jargon →
01

Servinetwork

A services company answered the same questions by hand every day, about documentation that already existed. I built a retrieval system that reads it and answers on its own — with explicit validation of every output, because a made-up answer costs more than no answer.

RoleDesign, backend and deploymentYear2024 — presentStatusIn production
Fig. 03 — Retrieval system architectureblue = the step that makes the difference
INGESTION · ONCEQUERY IN REAL TIMEMEASURED ON THE CLIENT'S REAL SUPPORT VOLUMEclientdocumentationchunkingembeddingsvectorindexuserquestionretrievalsemanticrerankingLLMoutputvalidationranks by real relevance,not by vector distanceif the source does not back it,it does not answer80% of queries resolved with no human involved
standard step critical design step— — — the index is built once and queried always
PythonLangChainOpenAI APIvector searchrerankingadvanced promptingAWS
02

Conversational agents · VentIA

The sales team handled every WhatsApp conversation by hand, including the ones that only asked for a price or a date. I designed an agent layer that classifies intent, queries the business systems and answers with real data — and hands the conversation back to a person the moment it stops being sure.

RoleAgent design, backend and integrationYear2025 — presentStatusIn production
Fig. 04 — Agent layer over the business systemsblue = where everything is decided
CONVERSATION ON WHATSAPPBUSINESS TOOLSMEASURED ON VENTIA'S CONVERSATION VOLUMEincomingmessageintentclassificationagentorchestratorreply to thecustomercatalogueand pricesquote buildervisitschedulinghands over toa persononly when the modelis not confident enough−60% of the sales team's manual handling
standard step critical design step— — — the escape path to a person
PythonFastAPILangChainMastraagent orchestrationWhatsApp Business APIPostgreSQL
03

Qore · attendance control

Attendance check-ins were being shared between coworkers: a photo of the code was enough. I built a system where the code is re-signed every fifteen seconds and every check-in validates the location against the site perimeter.

RoleProduct, backend and appYear2024StatusIn production
Fig. 05 — Attendance check-in with a rotating codeblue = the check that holds it up
EVERY 15 SECONDSATTENDANCE CONTROL SYSTEM IN OPERATIONthe serversigns a tokenthe screenshows the QRthe workerscans itvalid token+ geolocationattendancerecordedattemptrejecteda new one is signedevery 15 secondsthe token expires in 15 s;the radius is checkedagainst the siteoutside the radius or withan expired token, nothing15 seconds of validity per code, with the perimeter checked on every entry
standard step the double check— — — the attempt the system rejects
Next.jsTypeScriptPostgreSQLgeolocationJWT
Also built

ERP for a textile company

Production, inventory and invoicing lived in separate spreadsheets that nobody reconciled at month end. I built the system that unified them, with electronic invoicing built in.

Next.jsTypeScriptPostgreSQLelectronic invoicing

Sales forecasting

Purchasing was decided on intuition and inventory paid for it. I trained models on the real history to anticipate demand per product.

PythonXGBoostRandom Forestpandas
—
Renzo Lenes at his desk
Fig. 06Lima, Peru · 2026

Systems engineer, leading the development team at VentIA.

I started as a full stack developer and arrived at AI the long way round: first building the products, then realising that models are only useful when someone can use them without thinking about them. I do both today, and that is why my systems reach production.

I work remotely from Lima and take freelance projects when the problem is interesting. If you have a process that runs by hand and should not, I can probably help.

2026 — presentHead of Fullstack Development & AI EngineerVentIA · remote
2025 — 2026AI Engineer · RAG pipelines and agent orchestrationVentIA · remote
2025Data analyst · pipelines over financial dataENTEL
2024 — presentFreelance · management systems, RAG, electronic invoicingindependent
2026BSc Information Systems EngineeringUPC
Contact

Let's talk.

If you are building something with AI that has to work every day, or you have a manual process that has run out of road, write to me and we will look at it.

email
lenesrenzoalberto@gmail.com
whatsapp
+51 904 890 457
linkedin
renzo-alberto-lenes
github
RenzoLenes
cv
CV — AI Engineer (PDF)
CV — Full Stack (PDF)