I ship and automate production systems end to end. Payments, AI products, data, infra, usually alone, with agents doing much of the engineering.

SuyogParkhi

ScrollPayments · AI · Data · InfraAvailable 2026
(01) About

I run every layer of a consumer AI startup $290K ARR run rate and built payment‑data infrastructure for rails that carry half of India’s UPI traffic 10,000 TPS. I ship what the data allows, and kill the rest.

One engineer, every layer

Founding engineer at Sozo Labs, previously Juspay. Backend, infra, data, ML and mobile. Most of the engineering runs through agent harnesses, MCP servers over production and AI review on every diff, which is how one person covers a whole stack.

Bengaluru
IN
10K
01 · Peak, UPI

Transactions per second at the PSP where he built payment‑data infra.

$290K
02 · Run rate

ARR at Sozo Labs, from zero in the first three months.

2 MIN
03 · Deploys

Down from 15, with per‑PR previews and a Spot CI fleet.

400K
04 · Dataset

Messages embedded to test reply personalization.

(02) Selected systems

Systemsin motion(04)

Nothing here is a screenshot. Each figure is drawn live from what the system actually does: languages, transactions, sockets, queues.

01Sozo LabsFounding engineer · only engineer2025 — Now
Fig. 01 · 10+ Indian languagesLive

Sozo
Labs

Founding engineer · only engineer

A consumer AI startup with four products: an AI companion speaking 10+ Indian languages by chat and voice, a banking assistant on India’s Account Aggregator framework, an on‑device payment‑notification app and a market‑news app with realtime chat. He was the first engineer, left for Juspay, and came back to run all of it.

MRR
$24K
Deploy time
15 → 2 min
LLM spend
~$4K / month
Dup. reply race
7.1% → fixed
GrowthBookClickHouseLangfuseECSGemini
02Juspay · UPI DataProduct engineerDec 2025 — Jun 2026
0 txnsFig. 02 · 10,000 / s

Juspay
UPI Data

Product engineer

Payment‑data infrastructure at a PSP processing about 50% of India’s UPI traffic, inside NPCI‑regulated flows. He designed UPI‑Data‑OCD, a Nix and Terranix framework that turns ~1,000 lines of per‑tenant Terraform into 30–80 lines of typed Nix, and used it to stand up analytics for Kotak Bank and SuperMoney.

Terraform per tenant
1,000 → 30–80 lines
Tenant onboarding
days → 1 command
Long‑lived AWS keys
removed
HaskellKafkaCassandraClickHouseNix
03PerenniaRealtime infrastructure2026 · Solana
Fig. 03 · one snapshot, many socketsStreaming

Perennia

Realtime infrastructure · performance

A perpetual options venue on Pacifica, a Solana perp DEX, built with a quant collaborator who owned pricing. He built the realtime pipeline: Redis state snapshots behind a pub/sub layer, so WebSocket clients stream from cache instead of the venue recomputing state for every connection.

Risk engine
per‑account → bulk loop
Storage
local Docker → RDS + Redis
Concurrency bugs
fixed, with tests
PythonRedisPostgresWebSockets
04BingeClip AISDE internJan — May 2025
Fig. 04 · queue depth → workersKEDA

BingeClip
AI

SDE intern

The backend for a WhatsApp bot that generates AI video. Every user gets an isolated Redis queue, every client its own RabbitMQ workers, and KEDA scales the workers on queue depth, so one heavy user never stalls another.

Queues
isolated per user
Workers
per client
Scaling signal
queue depth
RedisRabbitMQKEDAKubernetes
(03) Decision log
Shipped, replaced, killed

The datadecides

    (04) LabCurrently studying · move to steer the field

    Opening up
    the model

    A structured mechanistic interpretability track alongside the day job, working toward a published replication and writeup. Before that, the applied side: embedding datasets, trained classifiers, pre‑registered experiments and eval design.

    1. 01 · TrackARENA
    2. 02 · ToolingTransformerLens
    3. 03 · FeaturesSparse autoencoders
    4. 04 · ControlGemma Scope steering
    5. 05 · GoalPublished replication
    (05) MethodHow one engineer covers a whole stack
    A

    Parallel research harnesses with specialised subagents and custom Claude Code skills, run as daily practice.

    JavaPythonRustHaskellTypeScriptC++SQL(Languages)
    KafkaClickHouseCassandraPostgresRedisRabbitMQ(Data)
    AWSKubernetesNixTerraformGeminiClaudeSolana(Infra · AI)
    (06) RecordWork and study

    2026

    • Sozo Labs

      Jul 2026 — Now

      Founding engineer, all technology · Bengaluru

    • Perennia

      2026

      Realtime infrastructure, with a quant collaborator · Solana

    • Juspay

      Dec 2025 — Jun 2026

      Product engineer, UPI Data · Bengaluru

    2025

    • Sozo Labs

      Jul — Dec 2025

      First engineer: payments, AI companion, DigitalOcean to AWS · Remote

    • BingeClip AI

      Jan — May 2025

      Software development engineer, intern · Remote

    2022

    • RCOEM

      2022 — 2026

      B.E. Computer Science, Shri Ramdeobaba College of Engineering and Management · Nagpur

    • IIT Madras

      2022 — 2025

      Diplomas in Programming and Data Science (BS Data Science) · Chennai

    (07) ContactFounding and early engineering roles

    Let’sbuild

    Write
    to me →
    suyogp0607@gmail.com
    LinkedIn ↗GitHub ↗
    © 2026 Suyog ParkhiBengaluru --:-- ISTDrawn in code, black and whiteBack to top ↑