press any key to skip
feed liveJPMORGANCHASE · SWE INTERN

VINEETSISTA

Low-latency systems · Quantitative development · AI products

I build systems that have to be fast — order books, ML pipelines, and AI products — and I care about the nanoseconds.

⌘K palette
VNT · limit order bookstreaming
Price · USDSize
187.7650
187.77360
187.78334
187.79417
187.80622
187.81912
187.82661
187.831,030
187.75spread0.02
187.7449
187.73242
187.72331
187.71411
187.70626
187.69905
187.68662
187.671,048
cumulative depth
session p&l+2.18%
price-time prioritymatching latency ~85 ns
// the operator

I'm Vineet — a CS Honors student at Ohio State who got a little obsessed with one question: how fast can a thing actually go?

Most days I'm somewhere between a C++ order book that argues about nanoseconds, a research lab teaching language models to explain themselves, and an AI product that has to ship intelligence before the market opens at 7:30.

I like problems that are equal parts fast, correct, and real. I build things that have to hold up when it counts — and I don't love waiting.

Let's build something that has to be fast.

vineet.sista@gmail.com
currently.log
  • interning
    JPMorganChase · Software Engineering
  • building
    A nanosecond C++ order book + an LLM inference engine
  • leading
    AWS Cloud Club @ Ohio State
  • researching
    Explainable medicine & clinical ML
  • reading the tape
    NASDAQ ITCH 5.0
  • based in
    Columbus, OH · from Naperville, IL
by the numbers
0K+
patient notes
cleaned & processed
+0%
malware accuracy
TF-IDF + RF tuning
0%
model accuracy
plant-care recommender
0ns
match latency
order-book hot path
0.0
GPA
honors · 4.0 scale
1 / 0
engineering scholars
competitive selection
01thesis

Vineet sits at the intersection of low-latency systems, quantitative finance, and AI products. The same instinct runs through everything — from a C++ order-book matching engine that cares about nanoseconds, to ML research probing how language models make clinical decisions, to shipping AI products that deliver intelligence on a deadline. He builds things that have to be fast, correct, and real.

FOCUS
Low-Latency · Quant · AI
BASE
Columbus, OH
STATUS
JPMorganChase SWE Intern
GPA
3.8 / 4.0
education
The Ohio State University
Honors B.S. in Computer Science
Aug 2024 → May 2028

Engineering Scholar — selected as 1 of 96 students for a competitive program focused on innovation and hands-on engineering projects.

relevant coursework
Data Structures & AlgorithmsSystems Programming / x86-64 AssemblyFiles & DatabasesObject-Oriented ProgrammingEngineering StatisticsLinear AlgebraElectronics
02position blotter

Experience

Eight positions, newest first — a trade history of where the work has been. Rows with detail expand.

  • Leads cloud architecture across the club’s AWS projects — compute, storage, data, security.
  • Mentors peers on system design: when to reach for which service, and why.
03watchlist

Instruments

Each project as a tradable instrument — ticker, thesis, live spec, and stack. Two flagship systems lead: a C++ order book and a from-scratch LLM inference engine.

VNT · composite index
187.42 +2.10%
OPEN
183.57
HIGH
194.08
LOW
177.21
watchlist · 9 instruments
time & sales
timepricesize
09:30:11.507187.24522
09:30:10.370187.27430
09:30:09.233187.29838
09:30:08.096187.31790
09:30:07.959187.33882
09:30:06.822187.33513
09:30:05.685187.34274
09:30:04.548187.35710
09:30:03.411187.34300
09:30:02.274187.37647
09:30:01.137187.39148
09:30:00.000187.42301
LOB● building
MATCH LAT~85ns
PRIORITYPRICE-TIME
STATUSBUILDING
C++20NASDAQ ITCH 5.0Zero-copy parsingLock-free replayMicrobenchmarking
VLLMsystems
KV CACHEPAGED
BATCHINGCONTINUOUS
APIOPENAI-COMPAT
PythonPyTorchTritonCUDAStreaming API
VELO
DELIVERY7:30 ET
CADENCEDAILY
ReactFlaskPostgreSQLStripeFinancial APIs
PHNTM
OUTPUT~2MIN VIDEO
PIPELINECLAUDE→11L→RMTN
Next.js 14FastAPICeleryRedisPostgreSQLClaudeElevenLabsRemotionDocker
WRTHlive
MODETACTICAL
STATUSLIVE
Next.js 14TypeScriptTailwind
ALPHA
SIGNALZ-SCORE
SIZINGFRAC KELLY
FlaskReactStatistical modeling
DRKM
SCORINGAI OPP
BRIEFINGDAILY
Next.jsTypeScriptLLM integration
MLWR
ACCURACY+16%
FEATURESTF-IDF
PythonScikit-LearnFlask
PLNT
MODEL98%
UISTREAMLIT
PythonScikit-LearnStreamlit
04signature · deep-dive

The Order Book Engine

A low-latency NASDAQ ITCH 5.0 limit order book engine in C++20 — plus a queue-position-aware market-replay backtester and a market maker with PnL / adverse-selection analytics. The flagship, in active development.

architecture · data path
FEED
ITCH 5.0 Feed Handler
Raw UDP/PCAP ingest of NASDAQ TotalView-ITCH 5.0 messages.
PARSE
Zero-Copy Parser
Zero-copy decode of Add/Cancel/Execute/Replace into the book.
BOOK
Limit Order Book
Full-depth reconstruction; intrusive, cache-conscious price levels; O(1) best-bid/ask.
MATCH
Matching Core
Price-time priority crossing; deterministic fills.
REPLAY
Market-Replay Backtester
Queue-position-aware replay; lock-free parallel across symbols.
MM
Market Maker
Quoting strategy with PnL & adverse-selection analytics.
BENCH
Microbench Harness
Hot-path latency probes; ~85 ns match, percentile capture.
MATCH LAT
~85 ns
PRIORITY
PRICE-TIME
LANG
C++20
REPLAY
LOCK-FREE
C++20NASDAQ ITCH 5.0Zero-copy parsingLock-free replayMarket makerMicrobenchmarking
interactive · cross the book● live
Price · USDSize
186.97113
186.98317
186.99412
187.00455
187.01919
187.02762
187.03780
186.96spread0.02
186.9590
186.94148
186.93418
186.92449
186.91910
186.90754
186.89778
hover a level to inspect resting orders →

Fire a market order and watch it walk the book — consuming resting size at price-time priority, just like the C++ core.

the optimization journey

How the hot path got to 85 nanoseconds

14.6×
faster than naive
00Naive std::map book
0 ns
01Flat array + intrusive lists
0 ns
02Cache-conscious layout
0 ns
03Branchless hot path
0 ns
04Zero-copy SIMD parse
0 ns
measured · p50 85 nsp99 ~140 nsper message · single core · -O3 -march=native
05ml systems · inference

miniVLLM

The second flagship — a from-scratch, high-performance LLM inference engine. Systems engineering all the way down to the GPU: paged KV cache, continuous batching, speculative decoding, a custom Triton kernel, and an OpenAI-compatible streaming server.

POST /v1/completions · streaming0/35 tok
promptdef is_prime(n):
throughput
0tok/s
ttft
19ms
paged kv-cache0/48 pages
continuous batching
req 1
req 2
req 3
req 4
KV CACHE
PAGED
BATCHING
CONTINUOUS
DECODE
SPECULATIVE
KERNEL
TRITON
PythonPyTorchTritonCUDA
06research lab

Explainable Medicine & Clinical ML

At OSU's BMBL and AIMed labs — probing how language models make clinical decisions, and mapping where urgent-care conditions cluster in latent space.

initializing webgl…
drag to orbit · scroll to zoom
sparse autoencoder · latent space
260K+
NOTES
de-identified patient notes processed (Regex + PyTorch)
5,000+
ACTIVATIONS
sparse-autoencoder activations trained
NER
PIPELINE
clinical entity extraction from raw notes
UMAP
VIZ
latent-space clustering of urgent-care conditions

Built an explainable-medicine workflow generating heatmaps by probing LLMs with targeted token removals — surfacing the features most predictive of clinical decision-making and reducing hallucinations. Trained a sparse autoencoder and used UMAP to visualize how urgent-care conditions cluster in latent space.

07system spec

Technical Arsenal

The datasheet — grouped by subsystem, the way a device spec or risk sheet reads.

$ cat system_spec.txtvineet@osu · rev a1f9c3e
LANG
languages
[10]
PythonC++JavaJavaScriptTypeScriptSQLRMATLABHTMLCSS
FRAMEWORKS
frameworks & libraries
[12]
FlaskReactNext.jsNode.jsPyTorchTensorFlowScikit-LearnNumPyPandasMatplotlibRemotionFramer Motion
INFRA
infrastructure & tools
[12]
GitDockerAWSPostgreSQLBigQueryAzure ML StudioRedisCeleryTableauLinuxStreamlitVS Code
all subsystems operational · 34 components loaded
08interactive · cli

Terminal

A real shell. Type a command — `help` to start. Arrow keys for history, Tab to complete.

vineet@terminal — bash
vineet.sh — interactive shell · type `help` for commands
vineet@osu:~
09open channel

The terminal is open.

Building something fast, correct, and real — or hiring someone who cares about the nanoseconds? Let’s talk.

or press ⌘K → copy email
vineet@osu:~$ COLUMBUS·OH · build a1f9c3e · © 2026 Vineet Sista