Abhinav Pangaria · Software Engineer · India

I build resilientsystems, from ideato production.

Full-stack engineer across backend, infrastructure and the product on top. Previously JP Morgan Chase. Stellar SCF grantee.

Open to roles
01 · About

I build things end to end: the idea, the system underneath it, and the part people actually touch.

The subject keeps changing. So far it has been developer tools, payments, privacy, language learning and research, sometimes alone and sometimes inside a large bank. The habit stays the same: understand the problem properly, build the smallest thing that really solves it, then keep at it until it holds up with real people using it.

Shipping: a web app
02 · Selected work

Thirteen systems, owned end to end.

01

Sieve

A vector search engine written from scratch in Rust: a billion vectors, searched on one laptop.

Vector SearchSole EngineerInteractive
113.5MPassages, One 24 GB Laptop
0.909R@10 at 4.8 ms p50
420 QPS1B Vectors, R@10 0.906
225 MBRAM, Everything on SSD
A billion vectors, searched on a laptop
A billion vectors, searched on a laptopHead to head with DiskANN on 12 threadsHow a query runs: route, scan, rerank, returnLatency at a billion vectors, against a 100 ms budgetIndex build time and memory, next to DiskANN
02

Voluntas

A Rust order-book collector for Kalshi and Polymarket that keeps every book exact.

Market Data InfrastructureSole EngineerInteractive
~550kPolymarket Tokens Held
41k to 43kUpdates per Second
118.2MSnapshots Replayed
0Mismatches
The result of two hours live on AWS
The result of two hours live on AWSEvery run, by host, and what each one foundHow correctness is proven: replay with zero mismatchesWhere the time goes, clock by clockKalshi, the second venue, through the same engine
03

Prefill and the Frame

Local inference vs. the browser compositor: why GPU-heavy pages stutter while a model reads.

Inference PerformanceIndependent ResearchInteractive
64%Dropped Frames, Batch 512
1.3%Dropped Frames, Batch 64
+9 to 11%TTFT Cost
717Trials Recorded
Findings
FindingsBreaking Chromium downWhat I'd do next
04

Loom

A shared, live world model for parallel coding agents, kept in one graph.

Agent CoordinationLead Architect & Engineer
0 / 36Merges Pass, Without Loom
24 / 36Merges Pass, With Loom
12 / 12Signature Changes
0 Double GrantsLease Race
05

Guardian Kane

The loop that makes an agent's "done" something it has to prove, not just say.

Verification SystemsLead Architect & Engineer
12 Numbered PhasesBuild
460+ Unit TestsTest Suite
3-Strike → HumanEscalation
4 Paired A/B BuildsMeasured On
06

AgentOps

The organizational control plane for agents that hold money, now on C# and .NET 8.

Agent GovernanceLead Architect & Engineer
C# · .NET 8, from TypeScriptStack
5 Agents, 2 ChainsAgent Org
Fail-closedPolicy Model
x402 · Permit2 · EscrowRails
07

Zidd

An AI English coach for learners in India, from Class 2 chapters to the interview room.

Adaptive LearningSole Builder
47NCERT Chapters Authored
2,330Backend Test Functions
36SQL Migrations
School + CareerProducts on One Engine
Landing: education that adapts to the learner
Landing: education that adapts to the learnerToonhub: each coach is a character with a practice stylePlacement diagnostic: snapshot, focused check, confirmationLearner report: wins, current focus and next challengeParent view: position, what changed, what needs attention