I build AI systems, backends, and the infrastructure behind them.
From a city’s 311 chatbot to a college digital library, I build tools that make information easier to find and systems easier to run.
Featured workretrieve
City-service information, in plain language.
I helped build a chatbot for Fort Wayne’s 311 service that answered residents’ questions using retrieval-augmented generation. A query cache handled repeated questions, while the knowledge base grew around the information residents asked for most.
question → cache
Corrected a validation error in LangChain's GPTCache integration when handling chat-generation results. PR #19427 was merged in March 2024.
From a design pattern to an illustrated tutorial.
Patternsmith turns software design patterns and code smells into tutorials with explanations, code, and diagrams. A staged pipeline develops and assembles each article, checks the output, and retries errors within set limits. If one topic fails, the rest of the batch can continue.
PipelineState ← topic
Course materials, in one searchable place.
I built Documan to bring scattered notes, lab manuals, and past papers into one library, available on the web and through Telegram.
Original platform
The Python/Flask system served 4,000+ students and delivered 200,000+ files before its retirement in May 2025.
Current rewrite
The Java/Angular rewrite adds four-tier access control and keeps search in sync with saved content. It is publicly available, with no users yet.
Design decisions
4,000+ students · 200,000+ deliveries · 1,200+ active users · 99%+ uptime
search being down never fails a write · public rewrite, no users yet
One configuration. Only the changes get deployed.
Hikari manages container deployments across machines from one configuration. It sends updates to an agent on each machine, which compares the new configuration with its last-applied state and updates only the affected stacks.
Configuration updated
Extract document data with local AI.
OAAM processes documents on the user's machine through a browser interface. A local vision model reads an invoice and returns structured data. An optional MCP integration sends extracted results to Google Sheets.
How it works
waiting for a file
See the route. Understand the effort.
Activity Visualizer turns GPX and TCX files into performance analysis, 3D route flyovers, and shareable videos. Review corrections to noisy readings before exploring the data, and use privacy zones to hide start and finish locations in maps and exports.
Activity files are processed in your browser. A Cloudflare Worker serves the app and proxies map tiles.
route replayillustrative · privacy zones at start and finish
See how your training changes over time.
Training Story turns Hevy workout exports into exercise trends, personal records, and comparisons. Review imported workouts, adjust how they count toward your progress, and explore every view through one shared analytics engine.
Workout files are processed in your browser.
weekly volumeillustrative · trend and records
Explore my worksearch
Choose a topic or enter a search to begin
Nothing matched. Try “backend”, “Redis” or “data visualization”.
↑ ↓ to move through results · esc to return to search
Browser-local keyword retrieval: each passage is a TF-IDF vector built when the site is compiled. Your query is analysed the same way, expanded with a short list of related terms, and ranked by cosine similarity over 43 passages. Results are grouped by project and ranked by their best passage. The map combines text similarity with category grouping; distances are illustrative.
Tools, systems, and collaborations
Academic projects
Select a technology to see where I’ve used it. Each number counts the projects and roles listed here.
AI & LLMs
Languages
Backend
Data
Infrastructure
Observability
Frontend
Testing
Experiencecontext
Northeastern University, Khoury College of Computer SciencesSep 2026 - present
Nebulogic Technologies, HyderabadJun 2023 - Jul 2025
Nebulogic Technologies, HyderabadJun 2022 - Jun 2023
Geethanjali College of Engineering and Technology, HyderabadDec 2021 - May 2023
Abouttokenize
I started Documan as an electronics engineering student who wanted a better way to share course materials. At Nebulogic, that interest grew into work on backend systems, deployment tools, and a city-service chatbot. Today, I’m studying computer science at Northeastern and teaching software engineering, while building projects across applied AI, infrastructure, and data visualization.
70 o200k tokens · 386 characters
Northeastern University
MS, Computer Science
Sep 2025 - May 2027 (expected)
Geethanjali College of Engineering and Technology
Bachelor's, Electronics and Communication Engineering
Aug 2019 - May 2023
Blogdecode
Latest post13 min readBuilding Reproducible Environments With Nix and NixOSThe four mechanisms Nix gives us for changing a system without losing track of it, declarative configuration, version control, ephemeral shells and NixOS specialisations, and how to pick between them.Contactgenerate