Creator, VISCA · Building in Bengaluru, India

I build verifiable AI for places that can't fail.

The trustworthy parts of frontier AI — systems that must be verifiable, autonomous, and safe in regulated, clinical, and physical environments.

Hospitals
Banks
Robots
Orbit
~90
services, one substrate
~30
namespaces
Azure regions, HA
solo
architected & built
Stanford Medicine/American Express/ETHIndia — Global Winner/Trust over IP/RISC-V International/MassChallenge/Clinton Global Initiative U./Stanford Medicine/American Express/ETHIndia — Global Winner/Trust over IP/RISC-V International/MassChallenge/Clinton Global Initiative U./

I work where the cost of being wrong is unacceptable.

I build the hard, trustworthy parts of frontier AI — systems that must be verifiable, autonomous, and safe in regulated, clinical, and physical environments. Shipped: VISCA, a self-maintaining regulated-AI platform; a clinical-AI product in pilot with Stanford Medicine; zero-knowledge identity systems (ETHIndia Global Winner). American Express Software and AI Engineer and published author on trust in AI agents. Currently also building a personal humanoid robot in public.

HospitalsClinical AI, in pilot inside Stanford Medicine — in-perimeter by design.
BanksEngineering trust at one of the world's largest financial institutions.
RobotsA personal humanoid built from scratch — actuation, perception, control.
OrbitA fault-tolerant runtime designed to reach satellite constellations.
Currently
  • CreatorVISCA
  • Co-creatorCare Collaborative · Stanford Medicine
  • Software Engineer IIAmerican Express · Global Servicing Org
  • Building1000 Days Robotics — a humanoid, in public
Education
Manipal Institute of Technology
B.Tech, Electrical & Electronics Engineering
2019 — 2023 · Manipal, India

What I'm building — and already running.

Independent initiatives — architected, built, and shipped hands-on. Real systems, in the real world, on infrastructure I designed from first principles.

Creator

VISCA

visca.ai · Bengaluru, India

The regulated AI stack that maintains itself.

Visit visca.ai
Kubernetes (k3s)Helm · Fleet GitOpsAzure ×3LiveKit / WebRTCNestJS · GraphQLTamper-evident audit
  • Building VISCA — the regulated AI stack that maintains itself: identity, scoped credentials, in-perimeter runtime, tamper-evident audit and the speech & language models, shipped inside the customer's perimeter (nothing leaves the wall).
  • Architected a multi-agent system on one governed substrate — real-time multimodal voice/video conversational agents, autonomous operator agents that deploy and maintain the platform itself, and post-session evaluation agents — multi-tenant and dynamically dispatched.
  • Designed and built it solo; runs as a self-maintaining, multi-region (3× Azure) HA Kubernetes platform (~90 services across ~30 namespaces) — humans only approve changes matching their security posture.
  • Targets the gap where ~88% of enterprise AI-agent pilots die at security review; first deployments in regulated healthcare.
Co-creator

Care Collaborative

Stanford Medicine · Remote

Clinical AI for rehearsing the hardest conversations in medicine.

Real-time avatarsStreaming ASR / TTSCompetency auto-gradingRuns on VISCA
  • Co-built a clinical AI that lets doctors rehearse difficult patient conversations with lifelike AI patients — real-time voice & video avatars that adapt to the clinician in the moment, then auto-grade empathy and clinical-communication competency.
  • Took it from zero to a pilot inside Stanford — clearing the institutional security & privacy review that kills most clinical AI — and built and operate it nearly solo on VISCA (my own stack).

Engineering at one of the world's largest financial institutions.

Engineering and security work inside regulated financial services.

American Express

Software Engineer II
APAC BISO → Global Servicing Org
Aug 2023 — Present
Bengaluru, India
  • Worked as a Software Engineer I in the APAC Business Information Security Office (BISO) — security reviews, risk metrics (KRIs), and regulator-facing work; later moved and promoted to Software Engineer II in the Global Servicing Org (Oct 2025).
  • Found and killed a zero-day in an internal service portal, and reported a separate flaw that could have exposed internal data across multiple systems — before either could be exploited.
  • Patent inventor (currently in filing); built GenAI proofs-of-concept in the Research group.
  • Published on the official American Express engineering blog — “Trust Without Disclosure: Why Zero-Knowledge Proofs Could Help Build Trust in AI Agents” (May 2026).

The long bets, built in the open.

Early, honest, work-in-progress. These are where I push hardest on the physical and the impossible.

Early R&D

OrbitalRuntime

Fault-Tolerant Runtime for Space-Based Datacenters

A fault-tolerant runtime that deploys and self-heals workloads across distributed, unreliable nodes — designed to extend to satellite constellations as the deployment substrate for space-based datacenters.

Build log coming soon

Backed by a track record of winning.

01Dec 2023

ETHIndia 2023 — Polygon ID ZK Track, Global Winner

Awarded by Polygon

Built a zero-knowledge identity layer on Polygon ID — verifiable, privacy-preserving credentials with selective disclosure for AI agents.

02May 2023

Stanford LLM × Law Hackathon — Finalist

Codex — Stanford Center for Legal Informatics

Built a GPT-4 + Unreal Engine 5 empathy-training platform for lawyers.

03Dec 2022

ETHIndia 2022 — EY Starlight Winner

Awarded by Ernst & Young

Built VISCA — Verifiable Interoperable Secure Cognitive Architecture — for medical data interoperability. The origin of today's company.

Also
MIT COVID-19 Hackathon Winner (×2)Smart India Hackathon FinalistAmex CodeStreet'21 — 1st Runner-UpTechgium FinalistSchneider Go Green India FinalistLinux Foundation OSS Japan Scholar
Fellowships & Programs
MassChallengeAlum (with Care Collaborative)
Clinton Global Initiative UniversityFellow
Y CombinatorStartup School Participant
a16z CryptoStartup School Participant
Standards & Open Source
Trust Over IP FoundationMay 2021 — Present
Contributor — Technology Stack Working Group

Help define technical standards, test suites, and interoperability / certification standards for the Trust-over-IP architecture stack.

RISC-V InternationalDec 2020 — Present
Contributor — Member

Contribute to the open RISC-V instruction set architecture.

From silicon to satellites — the full stack.

Languages, real-time AI, platform, infrastructure, security, and hardware. The breadth is the point: verifiable systems need every layer to hold.

01

Languages

Python · Go · Rust · Solidity

02

Real-time AI & Voice/Video Agents

Sub-second conversational voice/video agents (WebRTC) · low-latency LLM orchestration & multi-model routing · streaming speech-to-text (ASR), endpointing & speaker diarization · neural text-to-speech · real-time photoreal avatars (neural lip-sync) · VAD, turn-taking, barge-in & noise cancellation

03

Platform & Backend

NestJS · GraphQL · TypeORM · PostgreSQL · Redis · BullMQ · gRPC · React · Nx monorepo (Yarn) · generated client SDK

04

Infrastructure & Ops

Kubernetes (k3s) · Helm · Fleet GitOps · External Secrets + 1Password · Cloudflare (Tunnel & for SaaS) · Azure · S3 / MinIO · LiveKit egress recording · Prometheus / Grafana / Sentry

05

Security & Identity

In-perimeter / air-gapped deployment · tamper-evident audit · agent identity & scoped credentials · data-residency · zero-knowledge proofs · verifiable credentials · Cosmos SDK · CosmWASM · Hyperledger

06

Embodied / Hardware

Humanoid robotics · actuation & perception · IoT (LoRa, MQTT, M2M) · digital twins · edge compute

Contact

If you're building where failure isn't an option — let's talk.

Investors, founders, clinicians, and engineers welcome. The fastest way to reach me is email.

Pratyaksh Gupta · Bengaluru, India · onchainengineer.me