hescobariEST. 2009

Available · Q3 2026 · Two engagements open

One person.
The whole product.

Sixteen years writing, designing, deploying, and operating production systems across Bolivia, the United States, and Australia. Available for a small number of new engagements in 2026.

I'm Huascar Escobari. I take on the work that crosses design, code, infrastructure, and AI. The teams I join stop juggling four vendors and start shipping again.

Send a briefSee selected work
  • 0116+

    Years shipping production software

    First paying release in 2009, an attendance system for a school in Cochabamba.

  • 023

    Markets I've shipped in

    Bolivia, the United States, Australia. Three time zones, three currencies, one practice.

  • 0350K+

    Peak concurrent users

    On one platform, on a single weekday morning, on a database I architected and tuned.

  • 0440+

    Projects shipped end-to-end

    From the database schema to the typeface choice, written and operated by me or with a small team.

Selected work

/ 02

What I've shipped.

Four projects that demanded all four disciplines at once. Each one taught me something I couldn't have learned reading.

  1. 01 / Education platform

    SICE

    Bolivia · 2018–present

    Problem
    Schools across Bolivia were running grade tracking on WhatsApp screenshots and printed spreadsheets. No source of truth, no audit trail, parents asking teachers individually for every update.
    Approach
    Built a multi-tenant platform from a single-school pilot. Designed the data model so adding a new school took under an hour. Cached aggressively so 4G connections felt instant.
    Result
    40+ institutions onboarded, 18,000+ active students, peaks of 4,200 concurrent users during grade-release windows. Zero re-architectures of the core in seven years.

    Stack

    • Next.js
    • Postgres
    • Redis
    • Prisma
    • Fly.io
    Read the case study
  2. 02 / Logistics tooling

    Freight routing

    Australia · 2021–2023

    Problem
    A regional carrier was scheduling 60-truck routes in Excel. Re-routes during the day were manual phone calls. Drivers lost 90 minutes per shift to dispatch friction.
    Approach
    Built a dispatch-first UI on top of a rolling routing engine. Made the driver mobile view work offline so trucks didn't lose state in dead zones.
    Result
    One FTE reclaimed per quarter, 22% drop in driver idle time across nine months. The software paid for itself in week 14.

    Stack

    • TypeScript
    • Python
    • Postgres
    • OR-Tools
    • AWS
    • React Native
    Read the case study
  3. 03 / Financial operations

    Underwriting console

    United States · 2022–2024

    Problem
    Operators were copy-pasting fields between four internal tools to make a single risk decision. An ML model existed, but was a black box to the team using it.
    Approach
    Designed a single console with the model's scoring inline, every input editable, every output explainable. Full audit log on every operator override.
    Result
    Decision time dropped from 14 minutes to 4 minutes per case. Override rate cut in half. The operators asked the team to keep me on past the original contract.

    Stack

    • Next.js
    • FastAPI
    • Postgres
    • scikit-learn
    • Datadog
    Read the case study
  4. 04 / AI engineering

    Agentic workflows

    Remote · 2024–present

    Problem
    Teams were prototyping LLM agents in notebooks and shipping them with no evals. Production drift caught by users, not by the team that wrote the agent.
    Approach
    Built an evaluation harness, an observability layer, and an orchestration runtime that lets engineers write workflows in code, not config. Failures route to a human review queue.
    Result
    Three teams using it in production, 200K+ agent runs evaluated, two near-misses caught before they reached customers.

    Stack

    • Python
    • TypeScript
    • Anthropic
    • OpenAI
    • OpenTelemetry
    • Inngest
    Read the case study

Approach

Discover the work behind the work.

Read the field notes

«The systems that page someone at 3am are systems I've built, broken, and rebuilt. The point of design isn't aesthetics. It's that the user finishes the task without thinking about the tool.»

— field notes, somewhere between Sydney and Santa Cruz
  • 01Design
  • 02Engineering
  • 03Infrastructure
  • 04AI engineering

Five takes I've earned

  1. 01

    Boring middleware is the most underrated craft in our industry.

    Queues, retries, idempotency keys, ordered processing, dead-letter queues. The teams that get these right look like they have luck. They don't. They have invariants.

  2. 02

    Design and engineering are the same job at different layers.

    Pretending otherwise is how products get shaped like the org chart. The interface and the database both encode decisions about who the user is.

  3. 03

    AI isn't a layer above the stack. It's the new bottom of it.

    The engineers who already understand databases, queues, and observability ship LLM-powered systems faster than anyone arriving from the prompt side.

  4. 04

    Most incidents are inventory problems, not code problems.

    The team that knows what's running, where, and on what version, survives 3am. The team that doesn't, ships a runbook called 'restart everything'.

  5. 05

    Two weeks beats two months.

    If a build can't show real value to a real user inside two weeks, the scope is wrong. Slice smaller. Ship something. Re-scope from production data.

Services

/ 04

Three ways we work together.

Most engagements fit one of three shapes. Each has its own rhythm, deliverable, and end. Cash preferred, weekly or monthly retainer for ongoing work, fixed price for scoped builds. No hourly billing.

  1. Mode 01

    4 to 12 weeks

    Senior individual contributor

    Teams that need to ship faster without growing headcount.

    I join your engineering team as a senior IC. Pull requests, design reviews, code reviews, on-call rotation if it helps. I show up to standup, I write tests, I leave the codebase better than I found it. Two weeks of notice to end the engagement, on either side.

    • Daily availability inside your timezone overlap
    • Pull requests with the cadence your team expects
    • Design + code review on your existing work
    • Optional on-call rotation

  2. Mode 02

    3 to 12 months

    Technical lead, fractional

    Pre-Series A teams without a full-time engineering lead, or post-Series A teams during a search.

    Half the week, sometimes more. Architecture decisions, hiring loops, technical debt strategy, vendor evaluations, on-call calibration. I sit at the strategic table, but I still write code where it matters. Monthly check-in, quarterly written review.

    • Architecture and technical strategy
    • Hiring loops and interview design
    • Vendor and infra evaluations
    • Code in the path that needs the most senior hands

  3. Mode 03

    8 to 26 weeks

    End-to-end product build

    Founders shipping a v1, or teams replacing a contractor build that didn't land.

    From Figma to production, alone or with one collaborator. Database, API, frontend, infra, observability, runbooks, internal docs. Milestones biweekly, payments tied to them. The codebase I hand over is the codebase your next hire reads on day one.

    • Product design and user research
    • Database, API, and frontend implementation
    • Infrastructure, CI/CD, and observability
    • Runbooks and onboarding docs for your next hire

What I reach for

Tools earn their place by being honest about their failure modes. Here's what I default to, with the rationale.

  1. 01

    Languages

    TypeScript first, Python for ML and data, Go for infrastructure that needs to be small and fast. SQL fluent, not afraid of a query plan.

  2. 02

    Web

    Next.js with React Server Components. Tailwind for UI. tRPC or REST depending on the team. Prisma or Drizzle when an ORM is the right level of abstraction.

  3. 03

    Data

    Postgres for everything that fits. Redis for caches and queues. ClickHouse when Postgres isn't enough. SQLite for the dev loop and small embedded products.

  4. 04

    Infrastructure

    Vercel for the front edge. Fly.io, Railway, or AWS for the back. Terraform for what matters. Docker for everything else. Sentry plus OpenTelemetry for visibility.

  5. 05

    AI

    Anthropic and OpenAI for hosted reasoning, smaller open-weight models locally when latency or cost demands it. Inngest or temporal for orchestration. Custom evaluation harnesses; LangSmith only when the team is already on it.

Trajectory

/ 05

Three markets, one practice.

Each market taught me something the others couldn't. I've kept all three lessons.

  1. Bolivia

    2009 – present

    Where I started and where I keep building. Local product work, education systems, civic tools, the place where I learned that constraints are the best design teacher.

    Lesson: build for the network you have, not the network you wish you had.

  2. United States

    2019 – present

    Fintech, healthcare, internal tools at scale. Distributed teams across PST, EST, and CST. The market where I learned that boring software is what survives the regulator.

    Lesson: instrument first, optimize later, but instrument like you mean it.

  3. Australia

    2021 – 2023

    Logistics and operations tooling for a regional carrier. Long-haul async collaboration, Sydney working hours, the kind of remote work that only works if the writing is tight.

    Lesson: clear writing is faster than clear meetings, and ages better.

Often asked

/ 06

Before we talk.

01How do you charge?
Weekly or monthly retainer for ongoing work, fixed price for scoped builds. No hourly billing. Invoices in USD by default, EUR or AUD on request, paid net 7.
02What time zones do you work in?
I work from Bolivia (UTC−4), with regular overlap into PST, EST, GMT, CET, and AEDT. Most engagements have a daily two-to-four-hour synchronous window agreed up front. The rest is written.
03Do you take equity?
Cash is preferred. Equity considered for founding-engineer roles I would have joined as a hire. I won't sign for equity-only on a first engagement.
04Are NDAs okay?
Yes. Standard mutual NDAs are fine, send before our first call. I won't sign one-sided NDAs that prevent me from describing my own role at a high level on this page after the engagement ends.
05How do we start?
Email me with a sentence about the problem and a sentence about the timeline. If it's a fit, we schedule a 30-minute call. If we both want to move forward, you'll have a one-page proposal with milestones and price within two business days.
06What don't you do?
Crypto, gambling, ads-driven products without a clear value beyond impressions, fixed-bid death marches, and projects that need me to lie about timelines on a Tuesday and pretend to be surprised on a Friday.

Contact

/ 07

Currently open to a small number of new engagements.

Senior engineering, technical leadership, or focused builds. Two-week minimum on ongoing work, milestone-based for fixed builds. English or Spanish, any reasonable timezone, written-first collaboration.

Send a paragraph: who you are, what you're shipping, and roughly when. I read every email; I reply to most within two business days.

Writehuascarescobari@gmail.com
LinkedInGitHub