Niraj Dhakal — Baltimore, MD
Software engineer building production AI systems.
Go · React/TypeScript · PostgreSQL + pgvector · RAG architecture. Shipped multi-tenant infrastructure that real users \ndepend on — not coursework.
- 16+ mo
- production experience
- Multi-tenant
- AI platform shipped
- RAG
- pipeline built from scratch
Three builds, told as engineering stories.
Dogwood Gaming · Production
AI Marketing Platform
- problem
- A small game studio needed marketing content and competitor intelligence at a cadence no human team could sustain — across multiple tenants, each with their own data.
- built
- A multi-tenant platform from scratch: Go/Gin backend, React/TypeScript frontend, PostgreSQL with pgvector for semantic search, a Celery/Selenium scraping pipeline feeding competitor data, and LLM inference served locally through Ollama.
- outcome
- Shipped to real users with end-to-end ownership — architecture, debugging, deployment, and operations. Token-aware generation keeps inference costs predictable while surfacing posting-time and hook insights from scraped competitor activity.
- Go
- Gin
- React
- TypeScript
- PostgreSQL
- pgvector
- Celery
- Selenium
- Ollama
Retrieval pipeline
- 01IngestDocuments normalized and cleaned
- 02ChunkOverlap-aware splitting for context integrity
- 03EmbedVectors generated per chunk
- 04Indexhnswlib + SQLite approximate nearest neighbour
- 05Retrieve → ReasonAgent loops over retrieved context to answer
Personal project · Python
Agentic AI Research Tool
- problem
- Most 'AI research' tools are a thin wrapper over a chat API — they break the moment a question needs grounded, sourced context.
- built
- A research agent with a RAG pipeline written from the ground up: document ingestion, overlap-aware chunking, embedding generation, and vector indexing on hnswlib backed by SQLite — no managed vector service.
- outcome
- Answers grounded in retrieved source material, and a working understanding of every tradeoff in the retrieval stack rather than treating it as a black box.
- Python
- RAG
- Embeddings
- hnswlib
- SQLite
- Agents
Personal project · Systems
LocalStream
- problem
- Sharing a screen with someone nearby usually means a heavyweight account-gated meeting product routing your pixels through someone else's servers.
- built
- A peer-to-peer WebRTC screen-sharing app: create a room, share a six-character code, stream directly between browsers. Signalling, session negotiation, and reconnect handling built by hand.
- outcome
- Zero-install sharing that works in seconds, and hands-on experience with NAT traversal and real-time media. Currently being revamped into a full meeting app.
- WebRTC
- TypeScript
- P2P
- Signalling
- Real-time media
B.S. Computer Science
UMBC
I care about what happens beneath the API.
I'm a software engineer with a Computer Science degree from UMBC, but I've been building production software for over a year — not just coursework. At Dogwood Gaming I was one of a small team responsible for a multi-tenant AI marketing platform built from the ground up, owning large parts of the stack: a Go backend, a React/TypeScript frontend, PostgreSQL with pgvector, and an LLM inference pipeline on Ollama — along with the debugging, deployment, and operational reality of shipping software people rely on.
Rather than simply consuming AI APIs, I've built RAG pipelines, embedding-based retrieval, and the data ingestion systems underneath them. I like working from first principles and understanding the tradeoffs behind a technology instead of treating it as a black box.
That mindset extends past software. I design CAD models and 3D-print my own parts, and I enjoy taking things apart — literally and figuratively. Outside engineering I'm usually fishing or mountain biking, which is probably why one of my side projects is an AI-powered fishing app.
Currently looking for a new-grad software or AI engineering role — building products that ship to real users, alongside engineers I can learn from.
Real responsibility, early.
Internship → production ownership
Software Engineer · Dogwood Gaming
- Joined as one of a few students with no senior engineering oversight — the architecture decisions were ours to make and ours to defend.
- Built a multi-tenant AI marketing platform from scratch: Go/Gin services, React/TypeScript client, PostgreSQL + pgvector, Celery/Selenium ingestion.
- Owned the LLM inference pipeline on Ollama, including prompt/token budgeting and fallback behaviour when models or scrapes failed.
- Handled deployment and operations — the on-call reality of software that other people depend on.
University of Maryland, Baltimore County
Teaching Assistant / Teaching Fellow · UMBC
- Supported students through core CS coursework, debugging their programs alongside them rather than handing over answers.
- Sharpened the habit of explaining systems clearly — the same skill that makes code reviews and design docs land.
Tools I've shipped with.
- Go
- TypeScript
- Python
- SQL
- JavaScript
- React
- TypeScript
- Tailwind CSS
- WebRTC
- PostgreSQL
- pgvector
- RAG
- Embeddings
- hnswlib
- Ollama
- LLM inference
- Docker
- Celery
- Selenium
- Gin
- Linux
- Git
Open to new-grad software & AI engineering roles.
If you're hiring for a team that ships to real users, I'd like to hear about it. Fastest way to reach me is email.