Skip to content

About

Abdullah Azmi — Computational Genomics & Agentic Systems (CGAS) Lab & Engineering Portfolio at Northwestern University (abdullahazmi.com)

Resources

Stars

2 stars

Watchers

0 watching

Forks

Repository files navigation

Abdullah Azmi · CGAS Research Lab & Systems Engineering Portfolio (abdullahazmi.com)

Live Site CGAS Lab GitHub ORCID

Official researcher portfolio and Computational Genomics & Agentic Systems (CGAS) Lab portal for Abdullah Azmi at Northwestern University (Department of Computer Science & Northwestern Institute on Complex Systems — NICO).

Proposed Research Programme (6–12 Months)

The prospective research question is whether combining PaperPulse's source-grounded literature evidence with ChromAlphaGenome's regulatory-variant prioritization improves evidence traceability and prioritization against a fixed baseline. The planned first six months cover versioned data manifests, a benchmark, reproducible agent skills, and fixed-baseline versus agent-assisted evaluation; months 6–12 are for replication, error analysis, and preparing a methods submission. This is a plan, not a report of completed results. See the programme overview.


Featured Systems & Open-Source Repositories (28 Standalone Repositories)

I. CGAS Lab Flagship Scientific Packages (github.com/aazminu)

  1. aazminu/paper-pulse — Autonomous nightly scientific literature ingestion across ArXiv, Semantic Scholar, PubMed, and OpenAlex with hybrid PostgreSQL (pgvector) dense retrieval, Neo4j multi-hop citation-concept knowledge graphs, and claim-verified daily synthesis.
  2. aazminu/chrom-alphagenome — Multi-modal non-coding variant prioritization integrating AlphaGenome cis-regulatory tracks, ChromBPNet/DeepLIFT chromatin accessibility attribution, and MaxEntScan donor/acceptor splice-junction $\Delta\text{PSI}$ modeling across ClinVar, gnomAD v4, and ENCODE v4 cCREs (0.941 AUROC).
  3. aazminu/bayes-kalman-trace — Recursive Bayesian state-space filter and Rauch–Tung–Striebel (RTS) backward smoother across 500-sample noisy velocity trajectories with analytic missing-data marginalization and log-linear likelihood pooling over 250K measurements (93.8% held-out accuracy, 0.041 ECE).
  4. aazminu/fold-energetics-xtb — High-throughput AlphaFold 3 PAE spectral/connected-component rigid-domain partitioning, Foldseek 3Di structural homology search, and semiempirical quantum-chemical (GFN2-xTB) ligand pocket bioactivity triage.

II. Production Full-Stack, Data Operations & Open-Source Platforms (github.com/aazminu)

  1. aazminu/airtight-container — Production full-stack shipping container yard inventory, quoting, and idempotent month-end storage-and-handling billing platform (React 18, TypeScript, Node.js/Express, PostgreSQL, AWS EC2; 40K+ LOC, 25 migrations, 275+ automated tests).
  2. aazminu/LeadAI — Asynchronous batch entity research, multi-source web/API enrichment queue (pipeline_queue.py), and deterministic qualification scoring engine with exponential-backoff retry recovery.
  3. aazminu/odyssey — University computer science learning platform (Next.js, TypeScript, Strapi CMS, PostgreSQL) with recursive nested MDX lesson-preview rendering fixes, flicker-free code-editor state synchronization, and Jest regression suites.
  4. aazminu/mermaid — Declarative diagramming and charting library contributions (TypeScript, SVG/D3, Vitest, Cypress) supporting configurable inner/outer bar-chart data labels across horizontal and vertical orientations.
  5. aazminu/FarmData2 — Operational agricultural recordkeeping platform (Node.js, Vue.js, PostgreSQL, Docker) with transactional REST endpoints and deterministic Cypress E2E integration suites.

III. Systems Architecture, Distributed Data & Applied ML Suite (github.com/aazminu · PROFILE_PLAN.md)

  1. aazminu/data-cleaning-hackumass (ScrubData — Distributed Tabular Cleaning Engine) — Automated schema inference, composite deduplication, MAD/Z-score/IQR anomaly isolation, and reconciliation reporting across 100K+ row CSV/Parquet datasets (Python, Node.js, PostgreSQL, Spark/Databricks, S3).
  2. aazminu/python-riscv-simulator (Cycle-Accurate RISC-V RV32I Pipelined Simulator) — Bit-accurate single-cycle and 5-stage pipelined (IF, ID, EX, MEM, WB) RV32I processor simulator with EX/MEM and MEM/WB hazard forwarding, load-use stall bubble insertion, and golden-trace verification.
  3. aazminu/dividend-capture-spark (Market-Scale Dividend Capture & Streaming Alpha) — Distributed Spark/HDFS backtesting over 165K+ dividend declarations and 5 years of US equity bars paired with SPY-adjusted drop ratios, k-NN sector imputation, and temporal-split GBDT scoring.
  4. aazminu/world-model-compression-mpc (Low-Rank SVD World-Model Compression — LeWM) — Post-training uniform and activation-aware SVD (ASVD) spectral truncation and Hybrid Forward-KL + Elite-Classification distillation of latent world models for Cross-Entropy Method (CEM) Model Predictive Control on TwoRoom and PushT (1.72× rollout speedup, >96% reward retention).
  5. aazminu/adaptive-ml-moderation (Adaptive Real-Time ML Moderation & Drift Engine) — Sub-150ms p95 streaming toxicity inference combining hybrid word/char TF-IDF cascades and transformer scoring over 160K+ labeled records with leakage-safe splitting, priority triage queues, and Population Stability Index (PSI) drift monitoring.
  6. aazminu/agentic-scrape-repair (Agentic Scrape-Repair & Multi-Service ChatOps Bot) — Self-healing DOM selector repair engine gated on live schema verification paired with idempotent Python/FastAPI/Slack-to-ServiceNow/Rally/GitHub change synchronization (95% reduction in manual triage latency).
  7. aazminu/movie-recc (Real-Time WebSocket Group Recommender & Matrix Factorization) — Multi-user PyTorch collaborative filtering and residual bias-corrected group preference aggregation streaming live consensus rankings across concurrent FastAPI WebSocket rooms (<200ms latency).

IV. Additional Applied ML & Full-Stack Open-Source Repositories (github.com/aazminu)

  1. aazminu/sp500_platform — S&P 500 portfolio analytics, Sharpe/Sortino/drawdown evaluation, capped water-filling LLM conviction weight blending, and automated Alpaca sell-first rebalancing.
  2. aazminu/AP-Study-Guide-Generator — Automated AP exam scraper and sliding-window Retrieval-Augmented Generation (RAG) study guide generator with Recall@K, MRR, and nDCG@K evaluation.
  3. aazminu/Environmental-Disaster-Classifier — 10-year NASA EONET environmental disaster classification pipeline using 3D unit-sphere geospatial projection and inverse-frequency weighted k-NN.
  4. aazminu/grab-pic — Event photo sharing platform pairing a Spring Boot REST API with a GhostFaceNet + RetinaFace cosine-similarity selfie search worker.
  5. aazminu/comic-subscription — Marvel comic series tracker and chapter-delta email notification service.
  6. aazminu/motorcycle-rideability-index — Weather-driven motorcycle rideability index API and severe-weather risk classification service.
  7. aazminu/markov-music — Variable-order Markov chain musical sequence generator with lower-order suffix backoff, additive Laplace smoothing, log-sum-exp temperature decoding, and harmonic consonance evaluation.
  8. aazminu/NYC-Property-Scout — NYC real-estate geospatial BBL normalization, multi-source dataset discrepancy auditing, spatial Haversine comparable matching, and NOI/cap-rate valuation pipeline.
  9. aazminu/nitro-lab — Real-time Linux home-server and multi-container Docker health telemetry engine over Server-Sent Events (SSE) with online Welford/EWMA anomaly detection and SLA burn-rate calculation.
  10. aazminu/Watch-Engine — Hybrid Okapi BM25 and dense vector watch catalog search engine with Reciprocal Rank Fusion (RRF), MAD z-score price anomaly detection, and Recall@K/MRR/nDCG@K retrieval evaluation.
  11. aazminu/ClaimLess — AI medical-bill auditing and patient advocacy engine with GPCI locality-adjusted CMS Medicare fee-schedule benchmarking, NCCI unbundling detection, and settlement strategy synthesis.
  12. aazminu/finsim — Interactive historical market regime and portfolio simulator with CAGR/Sharpe/Sortino/VaR/CVaR risk analytics, cash-conserving SELL-first rebalancing, and seeded GBM Monte Carlo projections.

V. Verified Upstream Open-Source Pull Requests Merged (author:aazminu)

  • exmergo/dex#634 — fix(transform): rewrite ref() sites when renaming seeds and snapshots (#628): Resolved target node kinds via self.index.node_kind(name) in _Builder._matches (packages/dex-core/src/exmergo_dex_core/transform/propagate.py) and checked all _FILE_NODES in _refuse_surviving_reads so transform rename seed and transform rename snapshot rewrite downstream ref() call sites and YAML references atomically.
  • exmergo/dex#633 — fix(transform): preserve dbt-core 1.12 error cause behind level-tagged headers (#626): Updated _GENERIC_HEADER in _collect_messages (packages/dex-core/src/exmergo_dex_core/transform/build.py) to accept optional leading [LEVEL]: tags in dbt-core 1.12 JSON log headers so underlying error causes remain attached during shadow_parse and build.
  • GustavoBaranda/chatflow-agent#52 — fix: restore active agent and metadata when a turn rollback occurs: Checkpointed and restored session.active_agent and deep-copied session.metadata across both in-memory SessionContext and persistent SQLiteSessionContext (chatflow_sessions) when an unhandled exception triggers a turn rollback in Runner.run_async().
  • repowise-dev/repowise#3841 — fix(code-origin): recognise SwiftPM version-specific manifests as build files: Extended _BUILD_NAME_RE (packages/core/src/repowise/core/code_origin.py) to classify version-specific Swift Package Manager manifests (Package@swift-<version>.swift) as build files without misclassifying ordinary Swift sources.

Interactive Client-Side Workbench (#workbench)

The portfolio includes a zero-dependency client-side Interactive Scientific & Systems Workbench (assets/extension.js) with four live simulations:

  • BayesKalmanTrace State-Space Trajectory Smoother: Live 500-step constant-acceleration Kalman filter + Rauch–Tung–Striebel backward smoother with adjustable sensor dropout rate ($0%$–$45%$), observation noise $\sigma_v$, and log-linear pooling weight $\alpha$.
  • ChromAlphaGenome Variant & Splice-Disruption ($\Delta\text{PSI}$) Inspector: Base-resolution nucleotide $\Delta\log_2(\text{ATAC})$ accessibility attribution and MaxEntScan donor/acceptor splice shift explorer across ClinVar benchmark loci.
  • RISC-V (RV32I) 5-Stage Pipeline & Hazard Forwarding Simulator: Cycle-by-cycle pipeline stage stepper (IF · ID · EX · MEM · WB) demonstrating EX/MEM and MEM/WB forwarding vs. load-use stall bubble injection.
  • Airtight Container Idempotent Billing & Proration Verifier: Interactive PostgreSQL composite-key ledger demonstrating zero duplicate invoice generation across repeated month-end billing batches and webhook retries.

Local Preview & Verification

No build step is required to serve the static site. From the repository root:

python -m http.server 8000
node --test tests/portfolio.test.mjs

Then visit http://127.0.0.1:8000/ and http://127.0.0.1:8000/lab/.

About

Abdullah Azmi — Computational Genomics & Agentic Systems (CGAS) Lab & Engineering Portfolio at Northwestern University (abdullahazmi.com)

Resources

Stars

2 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages