IAGO PIRES

software engineer · brazil · UTC-4
j/k to scroll · ↑↑↓↓←→←→ba

Iago Pires software engineer portfolio

01 · HERO
> whoami
iago pires :: software engineer
> cat bio.txt
i build software. lately, software that talks back.
five years across frontend and backend, now focused
on ai in production: agents, eval workflows, and
integrations with external systems. typescript and
python in equal measure.
>
※ attention matrix · cursor-reactive

Skills and experience

02 · STATS · SKILLS · EXPERIENCE
5+
years shipping
4
companies
3
continents (br · de · us)
∞
shareholder value delivered
— skills —
core
typescript · python · javascript · react · next.js · fastapi · node
ai
agents · llms (gpt · claude · llama) · rag · eval & tracing · embeddings · vector search
infra
aws · lambda · kafka · docker · serverless · distributed systems
— experience —
2025 — NOW
Datagrid AI · Procore
Software Engineer · ai agents executing workflows across Procore / SharePoint / Notion · OAuth + tool orchestration · eval & tracing · acquired by Procore · SF (remote)
2023 — 2025
MVST.
Fullstack · Akina Health (€5M+ funded), RedCare pharma, Zurhost (70K+ DAU) · React/Next/RN · Munich (remote)
2023 — 2024
Motorola Research
AI in software testing · RAG + LLaMA pipeline for Android device QA
2022 — 2023
Freelance
crypto/metaverse SPAs · MERN inventory · serverless on AWS Lambda

Personal projects

03 · PERSONAL PROJECTS
01 quest
CLI · Agents
Local control plane for parallel coding agents . Plan, run, validate, integrate.
Go, JSON specs, multi-backend →
↳ github.com/iagopiresdev/quest
02 prompt-cache-lint
CLI · Lint
Static linter that catches Anthropic prompt-cache anti-patterns in CI, before they cost you money in production.
TypeScript, Anthropic API →
↳ github.com/iagopiresdev/prompt-cache-lint
03 spotter
Systems · Automation
Agent-reliability engine for adversarial web navigation. Every attempt leaves an evidence bundle; every failure lands in a 22-bucket taxonomy that drives deterministic retry policy.
Go, TypeScript, Python, Postgres →
↳ github.com/iagopiresdev/spotter

Writing

04 · WRITING (./blog/)
May 15, 2026 How we cut Claude API costs by 10x by understanding prompt caching A tour of the bugs, patches, and structural fixes we shipped to a forked Anthropic CLI tool to make prompt caching actually work.
→ all posts
※ tokens in, waves out

Local inference rig

05 · LOCAL INFERENCE (the rig)
qwen 3.6 27b · Q5_K_XL
a 3090 humming under the desk. building toward 4×3090 → 96gb vram for the bigger open models. instrumented end-to-end: agent-observability (otel → prometheus → grafana, dashboards as code).
slot 0
RTX 3090
24 GB
● ACTIVE · 22 tok/s
slot 1
[ wanted ]
RTX 3090
— goal —
slot 2
[ wanted ]
RTX 3090
— goal —
slot 3
[ wanted ]
RTX 3090
— goal —
22
tok/s
19.4 gb
vram used
32k
ctx window
61 °c
gpu temp
※ manual bench · live snapshots pending

Resume and contact links

06 · RESUME (./resume.pdf)
→ resume.pdf → resume-full.pdf → linkedin → github → email