CS @ UC San Diego '28 · GTM Engineer Intern @ GMI Cloud · ML Researcher @ Qualcomm Institute

Laksh Goyal

Applied ML systems, agent evals, and infrastructure for models that have to survive outside notebooks.

Open to new opportunities — internships, research, ambitious teams.
on rotation
spotify ↗
loading…
favorite films
letterboxd ↗
loading…
recently watched
diary ↗
loading…
01 — What I do

Build it, measure it, ship it.

I'm a CS student at UC San Diego who likes to build, break, and ship fast — mostly at the intersection of startups, medtech, and applied AI. Off the keyboard: films (Letterboxd), poker, F1 weekends, Lego, travel.

/01

Agentic AI & evals

Orchestration frameworks, eval harnesses, and self-correcting agents — measured on accuracy, cost, and latency, not vibes.

/02

ML systems

Vision and audio pipelines for messy real-world data — medical imagery, bioacoustics, RF signals, geospatial rasters.

/03

Automation & infra

Data pipelines, APIs, and dashboards that carry a model all the way from notebook to deployed, monitored product.

Currently: GTM Engineering @ GMI Cloud · ML research @ Qualcomm Institute (E4E, UCSD) — and always open to what's next.

02 — Stack

Tools I reach for.

Languages
Python TypeScript JavaScript Java C / C++ SQL
ML & Data
PyTorch scikit-learn LangGraph Anthropic / OpenAI APIs HuggingFace Ray Tune OpenCV MONAI LightGBM NumPy / pandas
Web & Backend
React / Next.js FastAPI Supabase / PostgreSQL Redis Node Tailwind
Infra & Tools
Docker AWS Vercel Cloudflare Modal GitHub Actions Azure ML Railway
03 — Work

Selected projects.

01 / 08

Pylon — Counter-UAS Sensor Mesh

Won the UCSD BowCapital Defense Hackathon; admitted to the Bow Capital incubator. Anomaly-based RF detection, dual SDR backends, UDP gossip mesh, 3D operator dashboard.

100% of novel emitters caught vs 0% baseline
Pythonscikit-learnNext.jsdeck.gl
visit ↗
02 / 08

Winnow — LLM Prompt Compression

Voice-first compression on two orthogonal axes: token-space (LLMLingua + self-implemented AttentionRAG behind a novel merge) and model-space (4-bit TurboQuant KV + LCLM).

~60× KV reduction, beats fp16 baseline F1
PyTorchModalDeepgram
repo ↗
03 / 08

AgentBench — Agent Eval Framework

4K-LOC framework scoring LangGraph agents on accuracy, cost, and latency percentiles — cost-adjusted accuracy metric, async-safe usage attribution, Ray Tune HPO, CLI + leaderboard.

n=55 · 0.964 accuracy · $0.042 total spend
PythonLangGraphFastAPI
repo ↗
04 / 08

Argus — Agentic Training Monitor

Autonomous agent polling live training logs every 10s; SPC detector across 4 failure modes dispatches bounded tool calls. Docker stack + live React dashboard on EC2.

49% of anomalies resolved autonomously
PyTorchAnthropic APIReactAWS
repo ↗
05 / 08

AgriShield — Wildfire Optimizer

7-stage geospatial pipeline turning farm polygons into ELMFIRE-ready rasters; 8-bearing fire-spread ensembles score 24 candidate firebreak layouts; no-build Canvas UI.

4.6K lines · 92 tests · mypy-strict clean
PythonrasterioELMFIRE
repo ↗
06 / 08

ChainSense — On-Chain Analytics

Ethereum pipeline ingesting 403K transactions and 371K ERC-20 transfers; 16-feature wallet matrix clustered into 9 behavioral archetypes with anomaly flagging.

59K wallets · 591 anomalies flagged
PythonRedisStreamlit
repo ↗
07 / 08

SitRep — Team PM App

Led an 11-engineer team through a 10-week Agile SDLC to ship a Scrum/Kanban/XP app on Cloudflare — auth, D1 migrations, AI standup digests, 5-stage CI.

11 engineers · 5 sprints · shipped live
JSCloudflarePlaywright
view live ↗
08 / 08

RepoMap — AI Codebase Visualizer

VS Code extension mapping any repo as an interactive dependency graph — multi-language parser (TS, JS, Python, Go, Java, Rust) plus AI file explanations and inline chat.

6 languages parsed · one-click AI edits
TypeScriptReactVS Code API
repo ↗
scroll to move the belt →
04 — Contact

Let's build something.

Open to new opportunities — internships, research collabs, hackathons, and ambitious teams. Fastest way to reach me is email.

lgoyal@ucsd.edu →