Sent emails (20)

Hi,

Getting an agent to actually operate instruments locked behind proprietary software with no real API is the kind of unglamorous, high-leverage problem I keep gravitating toward.

At Addie I've been the sole engineer building an AI voice platform from empty repo to production — LangGraph agents, WebRTC + OpenAI Realtime at sub-200ms, now serving students in 10+ countries. Before that I spent years in expensive-failure systems: ingesting tens of millions of 911 calls during the COVID surge, and developer infra at Twilio. I'm comfortable owning architecture, build-vs-buy, and on-call at once.

I work in Python/Rust/TypeScript, and the runtime-that-ties-it-together layer is exactly where I like to live — agents up top, low-level systems underneath.

Happy to walk through the voice platform architecture or anything else. Would love to chat.

Best,
Jose Oliveros
chinnno15@gmail.com

Hi Steven,

Kanary's cat-and-mouse framing resonates. I've spent 12 years building systems where failure is expensive, and lately shipping production AI end to end.

Something I built that was hard: at Addie, I was the sole engineer behind a real-time AI voice platform — LangGraph agents delivering counseling across web, SMS, voice, and WhatsApp at sub-200ms latency, 99.9% reliability across 10+ countries. The hard part was owning architecture, infra, model build-vs-buy, and on-call simultaneously with no fallback.

Earlier I built ingestion pipelines for tens of millions of 911 calls (Trino, Elasticsearch, S3) and scraping/cleaning pipelines feeding LLM agents that turn plain language into SQL. Python/Django, TypeScript/React, and LLM APIs are daily tools.

Keyword from your HN profile: [keyword].

Happy to share code or walk through the voice system on a call.

Jose Oliveros
chinnno15@gmail.com

Hi Jack,

I'm applying for the Backend + DevOps role. As founding engineer at Addie, I took an AI voice platform from empty repo to production serving students in 10+ countries — owning architecture, infrastructure, and on-call at once. That full-cycle, product-first ownership is what you're describing.

On the DevOps side, I've run AWS, Kubernetes, Terraform, and CI/CD in production, including geospatial systems where downtime was expensive. I'm SQL-first by habit and built data platforms on Trino, S3, and Postgres — so your stack (Trino, Postgres, Terraform, K8s) is familiar ground. Java is new to me, but I've shipped in Python, TypeScript, Rust, and Go; picking it up won't slow me down.

I write my own code, no Cursor. GitHub: chinnno15.github.io

Happy to walk through the architecture of any of these or chat whenever suits.

Best,
Jose Oliveros
chinnno15@gmail.com

Hi,

"Implementation is getting cheaper. Engineering judgment is not" is exactly how I've worked for the past year.

At Addie I'm the sole engineer behind a real-time AI voice platform serving students in 10+ countries at 99.9% message reliability. That meant owning ambiguous problems end to end: build-vs-buy across voice/transcription/model providers, cost-per-conversation targets, and architecture that scales past 100K hours annually — all under real reliability pressure.

Before that I built emergency-response systems for NIST/FEMA and developer platforms at Twilio, so multi-stakeholder constraints and compliance-heavy domains are familiar territory. Kyra's mix of healthcare, fintech, and compliance is the kind of complexity where judgment matters more than speed of typing.

Strong Python and TypeScript, US-based in Brentwood, CA.

Happy to share more or hop on a call whenever useful.

Jose Oliveros
chinnno15@gmail.com

Charles,

Your framing caught my attention: durable infrastructure for people who actually handle evidence, where failure isn't an option. That's the kind of work I've spent twelve years on.

At the International Public Safety Data Institute I built emergency-response systems for NIST and FEMA, ingesting tens of millions of 911 calls where data loss wasn't acceptable. More recently I architected an 8M-user crypto wallet handling private keys via Android KeyStore and iOS Secure Enclave — security as a baseline, not an afterthought. Since then I've been sole engineer taking AI products from empty repo to production.

The hardening, crash recovery, metadata preservation, and integrity work you describe is squarely where I operate. Local-first, no cloud trap, $50/month disrupting a bloated market — I like the thesis, and it's technically the right call.

Happy to share more or jump on a call.

Jose Oliveros
chinnno15@gmail.com

Hi Vilesh,

Your framing — idea → working web app → deployable product with auth, DB, payments, and custom domains — is exactly the kind of 0→1 build I've done before.

At Addie I was the first and only engineer, taking a real-time AI voice platform from empty repo to production for students in 10+ countries: LangGraph agents, WebRTC voice at sub-200ms, three Next.js products, and every build-vs-buy call along the way. At Twilio I worked on Yoyodyne, a codegen tool emitting snippets in six languages — directly relevant to reliable, maintainable AI-generated apps.

I own architecture, infra, product tradeoffs, and on-call at once, and I move fast without process for its own sake. Making AI-built apps debuggable and actually deployable is the interesting hard part.

Happy to share past work or jump on a call whenever suits you.

Best,
Jose Oliveros
chinnno15@gmail.com

Leo,

Your line about engineers who direct an army of agents but do the work themselves when needed describes how I already operate. At Addie I've been the sole engineer building a LangGraph agent system from empty repo to production—real-time voice on WebRTC + OpenAI Realtime at sub-200ms, serving students across 10+ countries at 99.9% reliability, while owning architecture, infra, build-vs-buy, and cost-per-conversation targets.

Before that I built LLM agents translating plain-language questions into SQL/Python against an ecommerce warehouse—the kind of AI-plus-knowledge work that maps directly to embedding procurement intelligence into real workflows for customers like Brex and Harvey.

I'm a full-stack generalist comfortable going wherever the hardest problem lives, and I've shipped systems where failure is expensive for 12 years.

Happy to share more or jump on a call whenever works.

Jose Oliveros
chinnno15@gmail.com

Hi Kinelo team,

Your framing of "context myopia" as workflow integration rather than model performance matches what I've spent the last year fighting. At Addie I was the sole engineer building a LangGraph agent system from empty repo to production — the hard part was never the model, it was giving agents the right context across web, SMS, voice, and WhatsApp reliably.

On how the software engineer's job is changing: we're shifting from writing code to designing the systems and context boundaries agents operate within — the judgment about what an agent needs to know, and when, is now the core work. That's exactly organizing humans and AI coworkers into shared workflows.

Twelve years shipping where failure is expensive (Twilio, emergency response, an 8M-user wallet), plus an M.S. in AI.

Happy to share more or chat.

Jose Oliveros
chinnno15@gmail.com

Hi,

Delta AI's scenario-based training platform is close to what I've spent the past year building. At Addie, I'm the sole engineer behind a real-time AI voice product: WebRTC + OpenAI Realtime API at sub-200ms latency, thousands of concurrent sessions, serving students in 10+ countries. That's the same conversational-AI-roleplayer problem you're solving for high-stakes practice.

On the full-stack side, I've shipped three Next.js products end to end with TypeScript, Node.js, and Postgres, owning API design, schemas, and architecture. I ship daily leaning on Claude Code, and I've built for expensive-failure domains before — emergency response for FEMA/NIST, an 8M-user crypto wallet.

I'm in the Bay Area (Brentwood) and comfortable with the founding-engineer scope: product, infra, and on-call at once.

Happy to share more or chat whenever works.

Best,
Jose Oliveros
chinnno15@gmail.com

Hi,

I'm applying for the Lead AI Engineer, Agentic Systems role.

At Addie I'm the sole engineer behind a LangGraph multi-agent system delivering real-time counseling across web, SMS, voice, and WhatsApp — sub-200ms WebRTC voice on the OpenAI Realtime API, students in 10+ countries, 99.9% message reliability. I owned the architecture, the build-vs-buy calls across model/voice/transcription providers, and cost-per-conversation targets.

Before that I spent years shipping systems where failure is expensive: emergency-response infrastructure ingesting tens of millions of 911 calls for NIST/FEMA, and developer platforms at Twilio. That regulated, high-stakes context is exactly where your finance/energy/healthcare clients live.

I've shipped Claude, OpenAI, and Gemini in production and treat evals as first-class, not an afterthought. Comfortable with monthly travel and embedding alongside client engineering teams.

Happy to share more or jump on a call.

Jose Oliveros
chinnno15@gmail.com

Hi Mercedes,

Your Forward Deployed Engineer role reads like the job I've been doing for the past year.

At Addie I was the only engineer, taking messy requirements straight from users and turning them into a LangGraph agent system delivering personalized counseling to students in 10+ countries. I owned everything: backend, infrastructure, on-call, and the product decisions in between — including when a deterministic tool beat a model call and when it didn't.

The jurisdiction-specific tooling you describe is familiar territory. At Rubercubic I built LLM agents that turned plain-language questions into SQL and Python against real ecommerce warehouses, mapping domain-specific rules into tools clients adopted as their primary interface.

I'm comfortable being the face of the team on client calls, and I ship fast — 100 hours of student conversations in month one.

I'm based in Europe. Happy to share more or chat.

Jose Oliveros

Hi Yair,

The founding engineer role is exactly the kind of work I do best: building AI products from an empty repo to production, owning architecture and infrastructure at once.

At Addie I'm the sole engineer behind a LangGraph agent system serving students across 10+ countries, built on the same stack you're using — TypeScript/Node, Python/FastAPI, React, Postgres. I've shipped production agents and made the build-vs-buy calls across model and tooling providers.

Your point about driving coding agents hard without losing the plot resonates. I lean on agents heavily but keep a real mental model of the whole system — twelve years across Twilio, emergency-response infrastructure, and an 8M-user wallet taught me that clean architecture matters most where failure is expensive.

Building a RAG context layer over all your tools is a problem I'd enjoy owning.

Happy to share more or chat.

Jose Oliveros
chinnno15@gmail.com

Hi Albert team,

You're building net-new consumer products from zero, with 20M+ users and real money on day one. That's the work I do best.

At Addie I was the first and only engineer — took an AI voice platform from empty repo to production, now serving students across 10+ countries at 99.9% reliability. Built real-time voice on WebRTC + OpenAI at sub-200ms across thousands of concurrent sessions, and owned every build-vs-buy call along the way.

Before that I was a core frontend engineer on a crypto wallet at 8M users, so scale and real consequences aren't new to me. Your stack (Python, Postgres, React/React Native) is where I live daily, and I ship prototypes on weekends because I can't help it.

Happy to share more or jump on a call whenever works.

Jose Oliveros
chinnno15@gmail.com

Hi Ryan,

Your posting resonates because I already work the way YC builds: no PMs, talk to users, decide, ship. As the sole engineer at Addie, I took an AI voice platform from empty repo to students in 10+ countries — architecting a LangGraph agent system across web, SMS, voice, and WhatsApp, with sub-200ms WebRTC voice at 99.9% reliability.

I'm already deep in the tools you mention — LLM APIs, coding agents, evals, context engineering — and I've owned build-vs-buy, cost-per-conversation, and on-call simultaneously. Earlier I built developer tooling at Twilio (docs platform for 250K accounts, code-gen across six languages), which maps closely to Work at a Startup and founder tools at scale.

I'm comfortable in Python, TypeScript, React, Postgres, and happy to pick up Rails fast.

Happy to share more or chat whenever works.

Best,
Jose Oliveros
chinnno15@gmail.com

Hi,

Your posting describes owning the intelligence layer end to end, working close to the infrastructure, and setting the engineering bar. That's exactly what I've been doing.

At Addie I'm the founding and only engineer. I built a LangGraph agent system delivering counseling across web, SMS, voice, and WhatsApp, real-time voice on WebRTC + OpenAI Realtime API at sub-200ms latency, and three Next.js products on top. I owned architecture, build-vs-buy across voice/transcription/model providers, cost-per-conversation targets, and on-call.

That's well beyond demos: thousands of concurrent sessions, students in 10+ countries, 99.9% reliability. Strong Python (FastAPI), and I've shipped multi-tenant systems in Docker/Kubernetes before.

The high-autonomy, no-layers setup partnering with a Product Owner on the how is how I already operate at early stage.

Happy to share more or jump on a call.

Jose Oliveros
chinnno15@gmail.com

Hi Senthil,

Sending this directly since you mentioned email is the best route.

Your stack — Python/FastAPI, Langchain, Postgres, Redis, GCP, K8 — maps almost exactly to what I've been shipping. As the sole engineer at Addie, I built a real-time voice AI platform end to end on FastAPI and LangGraph, took it from empty repo to production serving students across 10+ countries in month one, and owned the evaluation and cost-per-conversation targets that kept it honest.

The "rapid prototype to production, plus the evaluation that keeps them honest" line is exactly the loop I run daily. Twelve years across Twilio's developer platform, emergency-response infra, and an 8M-user wallet means I'm comfortable owning architecture and on-call at once.

Resume: chinnno15.github.io
LinkedIn: in my resume

Happy to share more or hop on a call whenever works.

Best,
Jose Oliveros
chinnno15@gmail.com

Hi,

I've spent the last year as the sole engineer behind Addie, an AI-first education platform — LangGraph agents delivering counseling across web, SMS, voice, and WhatsApp, plus real-time voice on WebRTC + OpenAI Realtime API at sub-200ms latency, now serving students in 10+ countries. That's production LLM work — agents, multi-channel delivery, cost-per-conversation targets — not wrappers.

Your posting mentions voice agents, multi-agent systems, and evals. That's exactly the territory I've been shipping in, and I default to Cursor/Claude Code to move faster. Owning a customer-facing feature idea-to-production without hand-holding is how I've worked as a founding engineer.

I'm US-based (Brentwood, CA), close to LA.

Happy to share more or jump on a call whenever works.

Best,
Jose Oliveros
chinnno15@gmail.com · chinnno15.github.io

Hi,

Your Senior AI Engineer opening lines up closely with what I've been doing. At Addie I'm the sole engineer behind an AI voice platform — I architected a LangGraph agent system delivering counseling across web, SMS, voice, and WhatsApp to students in 10+ countries, and built real-time voice on WebRTC + OpenAI Realtime API at sub-200ms latency.

Starbridge's zero-to-one work turning large-scale data into reliable sales insights is familiar territory. At Rubercubic I built LLM agents that turned plain-language questions into SQL, Python, and charts against an ecommerce warehouse, cutting a full analyst day to under 8 seconds. I've shipped deep document analysis, interactive chat, RAG, and multi-model deployments (OpenAI, Anthropic) in production, plus owned reliability and on-call.

Remote works, and I'm glad to be in NYC too. Happy to share more or chat whenever suits.

Best,
Jose Oliveros
chinnno15@gmail.com

Hi,

I've spent the last year as the sole engineer building Addie, an AI-first education product—LangGraph agents, real-time WebRTC voice, and three Next.js apps, all shipped from empty repo to students in 10+ countries at 99.9% reliability. That's the exact 0→1 work and stack DrSwarm runs on: Next.js + Django, Postgres, Celery/Redis, LLM APIs.

Before Addie I built emergency-response infrastructure ingesting tens of millions of 911 calls—systems where failure modes, retries, and reliability aren't optional. I've owned build-vs-buy calls, cost targets, and on-call at the same time, and I've worked directly with customers to shape what gets built.

Healthcare workflow automation (scheduling, RCM, patient ops) is a natural next step, and I overlap fully with US Pacific hours.

GitHub: chinnno15.github.io · LinkedIn on request. Happy to walk through Addie or the 911 warehouse—glad to chat whenever works.

Jose Oliveros

Hi Paul,

Saw the AI Product Engineer role on HackerNews. The description reads like my last year: at Addie I was the sole engineer who went from empty repo to a real-time AI voice platform serving students in 10+ countries — designing the UX, building three Next.js apps, and wiring up LangGraph agents across web, SMS, voice, and WhatsApp.

That work centered on the exact patterns you mention: human-in-the-loop counselor review, decision support, and AI-native interfaces beyond chat (sub-200ms WebRTC voice, not just a text box). I also owned build-vs-buy calls and cost-per-conversation targets — the kind of ownership real enterprise workflows demand.

Before that: LLM-to-SQL agents at Rubercubic, developer platforms at Twilio, and an M.S. in AI from UT Austin.

Happy to walk you through a demo or the architecture whenever works. 

Best,
Jose Oliveros
chinnno15@gmail.com