Data engineer · Full-stack developer · AI product engineerVisakhapatnam, India
I’m D L Narayana — a data engineer and full-stack developer building real-time CDC lakehouses, PySpark warehouses and AI-first web products. B.Tech Computer Science & Engineering · GITAM · Class of 2027.
01Selected work
Every card is a public repository with its own README, tests and, where it makes sense, a live deployment. All 17 projects →
01Data engineering
Real-time CDC lakehouse: Postgres → Debezium → Kafka → Spark → Bronze/Silver/Gold.
LakeFlow · architecture
Airflow · nightly backfill
02Data engineering
Schema-enforced PySpark batch warehouse over a million defective sales lines.
Retail Lakehouse ETL · architecture
Airflow DAG · JSON run metrics
03AI & computer vision
Multi-agent AI operations layer for autonomous drone fleets, with a human in the loop.
AeroSentry · architecture
JSONL traces · LangSmith optional
04AI & computer vision
On-device KYC: face match, passive liveness, OCR and tamper forensics — 100 % in the browser.
05Full-stack product
Full-stack Airbnb-style marketplace with a working booking flow and host analytics.
06Full-stack product
Civic platform: report city issues, upvote, and track fixes on a live dashboard and map.
02GitHub, live
Built from the GitHub API at build time — public repositories only, bot commits excluded. Snapshot 15 Sep 2026. Full table →
Public repositories
17
since Dec 2025
Commits, public repos
220
5 Apr 2026 → 4 Sep 2026
Active weeks
13 / 26
last 26 weeks
Languages
5
TypeScript leads at 65 %
Commit activity · 26 weeks · public repos
Language share · bytes across public repos
Recently pushed
Multi-agent AI operations layer for autonomous drone-in-a-box fleets: LangGraph supervisor + specialist agents, hybrid RAG, Pydantic tool calling, VLM/thermal analysis, human-in-the-loop safety gate, fallback protocol, tracing, golden-set evals, FastAPI, Docker, CI.
PythonPySpark batch ETL: 1M+ rows -> schema-enforced extract, harmonisation, dedupe, DQ quarantine gate, SCD2 star-schema warehouse, Spark SQL marts, MongoDB serving, Airflow DAG
PythonReal-time CDC lakehouse: Postgres -> Debezium -> Kafka -> Spark Structured Streaming -> Bronze/Silver/Gold with SCD2, data-quality quarantine, backfill and Airflow
PythonResumeForge — privacy-first career toolkit: live resume builder, real-time ATS scoring, PDF resume scanner, cover letters, interview prep, and application tracker. Pure HTML/CSS/JS, Apple design language, zero runtime deps.
JavaScriptClient-side document forgery detection lab — ELA, copy-move, noise, EXIF & JPEG forensics in the browser + PyTorch research pipeline
JavaScriptPrivacy-first in-browser KYC verification: face match, passive liveness, document OCR & tamper forensics — 100% client-side ONNX/WASM, zero inference APIs
Python03About
I’m a computer-science student in Visakhapatnam who would rather ship a working system than a slide about one. Over the last year that has meant end-to-end data pipelines — change-data-capture from PostgreSQL through Debezium and Kafka into Spark Structured Streaming, batch ETL into star-schema warehouses with SCD Type 2 history, data-quality gates, Airflow orchestration and idempotent, observable jobs.
It has also meant products: authentication, REST APIs, relational data models and polished front-ends — increasingly with AI at the core, from a LangGraph multi-agent operations layer with a human-in-the-loop safety gate to KYC and document forensics that run entirely on-device with ONNX and WASM.
Ask me about Spark internals (lazy evaluation, shuffles, broadcast joins, foreachBatch, checkpoints), Kafka and CDC semantics, dimensional modelling, data-quality strategy, SQL window functions, Next.js and React architecture, or shipping LLM features to production — RAG, agents, evals.
Replaying the same offsets or re-running a backfill yields identical tables. Retries are safe by construction, not by hope.
Every source is read with a declared StructType. Inference is only used to detect drift and log it before it becomes nulls.
One set of transforms powers both the stream and the batch path, so a single pytest suite covers both.
JSON structured logs, per-stage timers, quarantine breakdowns and run metrics files — observable by default.
04Stack
The ledger below is computed from every project’s declared stack — each filled cell is a repository you can open. “Exploring” means reading and running locally, not yet shipping.
| Technology | LakeFlow | Retail ETL | AeroSentry | VeriLens | StayNest | CityHelp | VeriDoc | DocuForge | Nova | ResumeForge | Stillpoint | GitHubLens | CodeRunner | AlgoViz | CryptoLab | Spectra | Typeflow | Σ |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Tailwind CSS (Web) | 9 | |||||||||||||||||
| TypeScript (Web) | 8 | |||||||||||||||||
| Motion (Web) | 6 | |||||||||||||||||
| Next.js (Web) | 5 | |||||||||||||||||
| React (Web) | 5 | |||||||||||||||||
| Vite (Web) | 5 | |||||||||||||||||
| JavaScript (Web) | 4 | |||||||||||||||||
| Python (Data) | 3 | |||||||||||||||||
| Airflow (Data) | 2 | |||||||||||||||||
| Docker (Data) | 2 | |||||||||||||||||
| ONNX Runtime Web (AI) | 2 | |||||||||||||||||
| Parquet (Data) | 2 | |||||||||||||||||
| Playwright (Tooling) | 2 | |||||||||||||||||
| PostgreSQL (Data) | 2 | |||||||||||||||||
| PySpark (Data) | 2 | |||||||||||||||||
| pytest (Data) | 2 | |||||||||||||||||
| PyTorch (AI) | 2 | |||||||||||||||||
| Recharts (Web) | 2 |
05Notes
Short engineering notes distilled from the project READMEs — the decision that mattered and the number that proved it. All 5 notes →
VeriDoc Studio2 min
VeriLens, VeriDoc Studio and DocuForge run detection, OCR and forensics client-side. What it took to make “the document never leaves the device” true.
AeroSentry2 min
AeroSentry lets AI agents schedule, inspect and alert on drone fleets — and stops every high-risk action at a deterministic policy check and a human approval.
VeriLens2 min
Quantising VeriLens cut the model payload by two thirds. Then the liveness model lost the ability to tell a face from a photo of one — so it stayed FP32.
06Timeline
Dates come from repository creation and push timestamps, not memory.
Sep 2026
Published LakeFlow (real-time CDC lakehouse) and Retail Lakehouse ETL (PySpark batch warehouse), plus AeroSentry — a multi-agent operations layer with a human-in-the-loop safety gate.
Aug 2026
A privacy-first career toolkit with a real-time ATS engine — pure HTML/CSS/JS, zero runtime dependencies, 36 commits over five days.
Jul 2026
VeriLens, VeriDoc Studio and DocuForge: KYC, Indian document verification and forgery forensics running entirely in the browser on ONNX/WASM.
Jun 2026
CityHelp, StayNest, AlgoViz, CryptoLab, GitHubLens, CodeRunner, Stillpoint, Typeflow, Spectra and Nova — each deployed live on Vercel.
Dec 2025
Account created; the public record starts here.
Class of 2027
Computer Science & Engineering, Class of 2027, Visakhapatnam. CGPA 7.97/10.
07Contact
Data engineering, full-stack or AI product work — internships, roles, collaborations. I read every message and reply from nvr0910@gmail.com.