Feed raw lyrics → get musically coherent chord progressions. DistilBERT classifies emotional mood, Mistral 7B generates contextually appropriate chords, and MuseScore renders final sheet music.
92.4%
Mood Accuracy
4.3/5
Chord Coherence
~1.8s
Inference / song
DistilBERT
Mistral 7B
LangChain
HuggingFace
FastAPI
MuseScore
Simulates a parallel file system behind a Kubernetes CSI workflow — separating metadata (MDS) from block data (OST). Includes fault injection, Prometheus metrics, Grafana dashboards, fio benchmarking, and a live demo UI.
5
MDS Operations
4
OST Operations
3
Fault Injectors
Go
gRPC
BoltDB
Kubernetes
Prometheus
Grafana
Docker
fio
End-to-end medallion ETL on 75M+ NYC taxi records. S3 → Glue → Lambda → Snowflake with automated partitioning, data quality checks, and optimized columnar storage.
75M+
Records Processed
35%
Query Speedup
↓28%
Pipeline Cost
AWS Glue
Lambda
Snowflake
S3
Python
Medallion
Cooperative multi-drone navigation using joint policy learning. Benchmarks Q-Learning, DQN, and QMIX on a custom PettingZoo environment with up to 8 simultaneous agents.
+20%
Faster Convergence
↓38%
Reward Variance
8
Max Agents
PyTorch
QMIX
PettingZoo
DQN
Python
Dual-layer RoBERTa verifier wraps GPT outputs — claim extraction, NLI scoring, confidence gating. Reduces hallucinations 40% over raw GPT-3.5 with only 210ms overhead.
86.3%
Detection Accuracy
↓40%
False Outputs
+210ms
Avg Overhead
RoBERTa
OpenAI
HuggingFace
NLI
Python
Classifies 14,640 airline tweets into Positive/Neutral/Negative with sarcasm tone detection. Addresses severe class imbalance via SMOTE, then deploys MultinomialNB with real-time prediction input.
84.6%
Accuracy
77.6%
F1-Score
14.6K
Tweets Analyzed
scikit-learn
MultinomialNB
SMOTE
TF-IDF
NLTK
Python
40+ technical indicators engineered for signal generation. Optuna HPO + MLflow experiment tracking slashed iteration cycles by 20%. Backtesting suite in progress.
Documentation & Proof of Work
75%
40+
Indicators
+20%
Faster HPO
···
Backtest P&L
⏳
Proof of Work Loading
Backtest results, equity curves & write-up being added
MLflow
Optuna
Python
TA-Lib
Pandas
Full kernel implementation: threads, virtual memory, filesystem, and system calls. 5K+ lines of C + x86 assembly with priority donation scheduling and MLFQ.
Documentation & Proof of Work
40%
5K+
Lines of Code
90%+
Test Coverage
···
Test Suite SS
⏳
Proof of Work Loading
Test suite results, screenshots & design doc being added
C
x86 Assembly
OS Design
MLFQ
Virtual Memory
Production-ready collaborative text editor — multiple users edit simultaneously with CRDT conflict-free merges, multi-cursor presence, offline-first IndexedDB sync, and role-based sharing (Owner/Editor/Viewer).
0
Merge Conflicts
2s
Autosave Interval
7
Core Features
Next.js 15
React 19
TypeScript
TipTap
Yjs CRDT
WebSocket
Prisma
NextAuth
terminal
$ git checkout -b next-fullstack-project
Switched to new branch
$ npm run build
▶ Compiling... 0 errors
$ git push origin main # coming soon
█
⚡
Production SaaS applications
🔐
Auth, REST & GraphQL APIs
☁️
Cloud-deployed with CI/CD