I build AI systems that survive contact with production—

governed, observable, and made to scale.

Haris Jalal

Applied AI / San Francisco
LinkedIn ↗ GitHub ↗
Scroll to explore ↓
(01) Personal operating system

Builder by instinct.
Reviewer by habit.

I’m at my best where ambitious AI ideas meet real-world constraints—and where clear thinking matters as much as clean code.

Build.

Blank page to production

I turn ambitious ideas into working systems across product, platform, and infrastructure.

Think.

Systems over features

I connect models, software, teams, and failure modes before they become surprises.

Prove.

Evidence before confidence

I test assumptions, review claims, and make quality visible and measurable.

Finish.

The last mile matters

I care about the unglamorous details: reliability, clarity, adoption, and what happens after launch.

(02) Point of view

Anyone can make an agent look smart in a demo. I make it dependable.

I work where orchestration, retrieval, evaluation, safety, and infrastructure meet—turning brittle prototypes into platforms teams can trust.

(03) Selected work

Proof over promises.

Three independent builds. Each one designed around evidence, guardrails, and real-world failure modes.

01 / Research agents Antler Hackathon

AutoArena

Autonomous adversarial research with evidence-gated experiments, deterministic replay, and immutable leaderboard receipts.

  • 44 → 69%strategy win rate
  • 112held-out evaluations
  • Fulltrace + lineage export
02 / Voice agents Y Combinator Hackathon

Better Call Ron

A real-time voice agent for caller triage and live attorney transfer, with signed webhooks and policy-bounded tools.

  • 94%end-to-end success
  • 74 secmedian completion
  • 50full trials
03 / Embodied AI Y Combinator Hackathon

Tarry

An embodied multimodal memory agent that turns live audio and whiteboards into durable, queryable office memory.

  • 90%top-1 retrieval
  • 120memory queries
  • Multiaudio + vision
(04) Experience

Built across the whole stack.

From product foundations to governed agent platforms.

01

Feb 2025 — Present

Senior Software Engineer

Oracle / AI Agent Platform

Built a governed MCP/A2A platform, a LangGraph runtime for 20+ assistants, and automated release gates from code to Kubernetes.

  • MCP / A2A
  • LangGraph
  • Kubernetes
  • Redis Streams
02

Jul 2022 — Jan 2025

Software Engineer

Oracle / Conversational AI

Built the platform behind 50K+ monthly conversations, a Preact SDK used by 18 applications, and an ingestion plane spanning 2.8M objects.

  • Preact / TypeScript
  • Java 21
  • Helidon
  • WebSockets
03

Apr 2021 — Sep 2021

ML Research Engineer

Drexel University

Developed forecasting models that reduced error below 5% across seven product lines, outperforming manual planning baselines.

  • scikit-learn
  • Prophet
  • Forecasting
  • Evaluation
04

2019 — 2020

Early systems work

Oracle + Merck

Shipped React product features, improved Electron startup performance, managed enterprise requirements, and automated reporting.

  • React
  • Electron
  • Salesforce
  • Product delivery
(05) About / capabilities

I like the hard part after “it works.”

I’m Haris, an applied AI engineer in San Francisco. For 4+ years I’ve built the infrastructure that moves AI from promising prototype to dependable product.

My range is deliberately broad: agent orchestration, enterprise retrieval, evaluation, safety, backend systems, frontend SDKs, and the operational layer connecting them.

Peer reviewer for NeurIPS 2026 and COLM 2026.

01

Agents & orchestration

LangGraph, MCP, A2A, tool calling, OpenAI Realtime

02

Retrieval & evaluation

RAG, Vector Search, reranking, embeddings, LLM evaluation

03

Platforms & infrastructure

FastAPI, Java, Redis, Kubernetes, OCI, Docker, CI/CD

04

Safety & product

Guardrails, PII controls, observability, React, TypeScript

(06) Contact
harisjalal502@gmail.com ↗

Let’s make something that lasts.

San Francisco, CA

Haris Jalal.