01 / Origin
A histone protein from a real crystal structure (PDB 1KX5), the DNA wrapped around it, and a helix whose base pairs spell my name.
Anish Nehete
Data Scientist and Engineer
Clinical NLP, speech, geospatial and data systems.
02 / RAVEN · USC Keck School of Medicine
A rule-based clinical NLP pipeline that screens pathology reports for cancer-registry reportability, with every decision traced to its rule and sentence.
Synthetic example, not patient data.
REPORTABLE
Invasive ductal carcinoma is affirmed in the final diagnosis and passes every rule.
98.4% agreement with final registry decisions across 2,193 pathology reports.
An agreement figure, not diagnostic accuracy. RAVEN screens pathology reports for cancer-registry reportability; it does not diagnose cancer.
The pipeline: HL7 v2 parsing, sections and specimens, medSpaCy terms, ConText for negation and history, SEER and California rules, and a trace for every decision.
On screen
Both cases are synthetic reports, not patient data.
Choose a mention or a rule to see its reasoning.
The record One reportable mention makes the whole record reportable.
SYNTHETIC PATHOLOGY REPORT
FINAL DIAGNOSIS
A. BREAST, LEFT, CORE BIOPSY:
INVASIVE DUCTAL CARCINOMA, GRADE 2.
ADDENDUM
B. LYMPH NODE, SENTINEL, EXCISION:
NO EVIDENCE OF METASTATIC CARCINOMA.
CLINICAL HISTORY
HISTORY OF MELANOMA, LEFT ARM.
The synthetic message as a laboratory system writes it: HL7 version 2 segments, one per line, fields separated by vertical bars. Not patient data.
MSH|^~\&|SYNTHETIC|PATHOLOGY PID|1|||SYNTHETIC^RECORD PV1|1|O|SYNTHETIC OBX|1|TX|FINAL DIAGNOSIS A. BREAST, LEFT, CORE BIOPSY: INVASIVE DUCTAL CARCINOMA, GRADE 2. OBX|2|TX|ADDENDUM B. LYMPH NODE, SENTINEL, EXCISION: NO EVIDENCE OF METASTATIC CARCINOMA. OBX|3|TX|CLINICAL HISTORY HISTORY OF MELANOMA, LEFT ARM. SIGNED: SYNTHETIC PATHOLOGIST\.br\
In the scene: point at a phrase in its port to see how it was read; press it to follow its route.
03 / Intelligence · Pasadena Digital
An ocean of radio static, and one synthetic ATC phrase rising out of it as Whisper hears it: 80 mel bands, word by word.
Whisper-Medium fine-tuned on 12K+ hours of noisy air traffic control audio.
Hyperparameter tuning and RMS-based silence filtering on the training audio.
On screen
The phrase is synthetic. Transcription only: nothing here directs an aircraft.
04 / Systems
One machine, three systems: an airport graph, a natural-language query console and a fleet of coordinated agents.
OwlVision: 95% of redundant taxiway paths pruned by cycle detection, centrality analysis and pruning on OpenStreetMap data.
ChatDB: natural-language questions translated to SQL and MongoDB queries.
Multi-agent trading: up to 8 LLM agents per cycle over 6 MCP servers.
On screen
The graph is illustrative, not airport data, and the query is a sample on an illustrative schema.
05 / Matrix
The graph's routes rise off the plaza as their own names, and the towers stream its routing data as light.
06 / The Sphere
A real-time geospatial intelligence platform: signals normalised, deduplicated and placed on the globe.
Ingestion: 9 signal adapters polled on 60-second cycles with asyncio. Each signal is normalized and deduplicated, and a failing adapter is isolated from the others.
Services: retrieval, portfolio and alerts as near-real-time microservices (FastAPI, Next.js) over REST and WebSockets.
State: PostgreSQL with Alembic migrations, and Redis rate limits.
Docker Compose with health checks, tested with pytest, Vitest and Playwright.
On screen
The Earth is NASA's Blue Marble and Black Marble imagery.
The signals, the events and the investigation are sample data, not a live feed. The satellites are illustrative motion, not data the platform ingests.
07 / The people who made it
Photographs with friends, set large into a room of flowers and data.
08 / Through my eyes
My own photographs of places and training, held in light. Point at one to bring it closer.
09 / Human
Mount Fuji, rebuilt in 3D around a photograph of mine. Two of my papers are in its sky: point at or select the birds or the bright star.
Choose
Arrow keys move the probe, Enter interacts where it points, Escape leaves.
Loading the worlds
After the story
Clinical registries, air traffic speech, geospatial intelligence and data platforms: the systems, what I built, and what changed.
98.4%
agreement with final registry decisions
RAVEN, USC Keck School of Medicine
612
tables and charts built in under a minute
SEER*Stat pipeline, USC Keck
12K+
hours of noisy ATC audio used to fine-tune Whisper-Medium
Whisper-Medium, Pasadena Digital
95%
of redundant taxiway paths pruned
OwlVision, Pasadena Digital
50%
less employee onboarding time
Onboarding tool, Plural Technology
60%
less manual labeling
VLM annotation tool, Pasadena Digital
9
signal adapters on 60-second ingest cycles
The Sphere, geospatial platform
to Present
University of Southern California, Keck School of Medicine
Cancer Surveillance Program
Los Angeles, CA
to
Pasadena Digital, USC Viterbi Startup Garage
Los Angeles, CA
to
Plural Technology
Pune, India
to
ByteOrange
Pune, India
to
A full-stack, real-time geospatial intelligence platform. Microservices handle retrieval, portfolio and alerts, fed by ingestion from 9 signal adapters.
to
A hybrid recommender built in PySpark on raw RDDs. It blends item-item collaborative filtering, matrix factorization and XGBoost into one ensemble.
to
A multi-agent system in which LLM agents research markets, execute validated trades and log their rationales through MCP servers.
to
A natural-language interface for SQL and NoSQL databases. Plain-English questions become schema-grounded MySQL and MongoDB queries.
IEEE WCONF
Implemented PPO and SAC reinforcement learning models, tuning reward functions and policy and loss dynamics to train adaptive non-player character (NPC) behaviors.
IEEE ASIACON
Compared YOLOv5 and YOLOv8 in a detection pipeline that identifies rare bird species for live biodiversity monitoring.
ICDAI
Trained artificial neural network and XGBoost models on NASA exoplanet data to detect exoplanets from light-curve patterns and predict habitability.
University of Southern California
GPA: 3.48/4
Los Angeles, CA
Maharashtra Institute of Technology, WPU
GPA: 8.62/10
India
I do fitness training and use Strava.
If this is relevant to what you are building, get in touch.
