Malco Salcedo

A Raspberry Pi box that listens to a CD player, identifies what's playing via audio fingerprinting, and shows the collection — plus what's currently spinning — to friends on a live web page.

• Built a catalog service where albums are added by barcode, resolved against the Discogs and MusicBrainz APIs for metadata, cover art, and tracklists.
• Built a now-playing pipeline that samples audio from a mic/line-in, fingerprints it with Chromaprint (fpcalc), and identifies the track via the AcoustID API — constrained to the selected album's known tracklist, with a local fingerprint cache and silence-gap detection to keep matches fast.
• Designed the SQLite schema and FastAPI web app tying the catalog, now-playing state, and listen history together, running entirely on-device.
• Planned public sharing via a Cloudflare Tunnel, exposing the live now-playing page with no port forwarding or separate server.

A binary classification model trained on a 105,000-sample geospatial dataset to predict wildfire risk, paired with a live decision-support map.

• Trained a binary classification model achieving 90.2% accuracy and 84.3% recall.
• Engineered 16 features from NASA satellite and sensor data — NDVI, evapotranspiration, precipitation deficit, soil moisture proxies, and land surface temperature.
• Tuned a cost-sensitive classifier using spatial covariates (terrain slope, road distance, population density) to minimize false negatives.
• Deployed a visual decision-support interface with a live geospatial risk map and auto-generated operational reports.

An Electron desktop app that auto-classifies drum one-shots (kicks, snares, hi-hats, claps, toms, 808s, cymbals) so producers can organize sample libraries without hand-tagging every file.

• Built the librosa feature-extraction pipeline, deriving spectral centroid, bandwidth, contrast, and flatness, MFCCs, ZCR, and RMS statistics from each audio sample.
• Trained and evaluated a random forest classifier over those acoustic features to auto-label drum one-shots by type.
• Bundled the trained model into the app's Flask backend so classification runs locally without a network round-trip.

A full-stack career assessment platform built with UCSC Blueprint for Your Future Is Our Business, a Santa Cruz County nonprofit, helping K-12 students explore careers and connect with local job opportunities.

• Built the student-facing quiz engine, scoring multi-dimensional responses against a weighted career-matching algorithm to generate personalized recommendations.
• Implemented the career exploration flow in React/Next.js, mapping quiz results to browsable career profiles tied to local Santa Cruz County job opportunities.
• Developed an admin dashboard with Recharts-powered visualizations for tracking student engagement and quiz completion across schools.
• Integrated Firebase (Firestore + Admin SDK) as the backing data layer, wiring real-time reads/writes between the quiz engine, admin dashboard, and reporting views.