Nikunj Patel
Backend AI Engineer
I build production RAG systems and the backends that keep them fast and reliable. Three years shipping GenAI, Flask, and cloud systems at Pragnakalp Techlabs.
Claude Certified Architect, Foundation · Anthropic(opens in a new tab)
70-80%
48
4
20 km
Projects
Selected work
01
No-Code RAG Chatbot Platform
Turns documents and websites into deployable chatbots for web, Telegram, and Slack.
- FastAPI
- LangChain
- OpenAI
- Milvus
- Documents and website pages are embedded into Milvus
- A user asks the bot on the web, Telegram, or Slack
- The backend retrieves matching chunks and answers through the same channel
02
AI Security Scan-and-Remediate Pipeline
Strix finds and validates exploits, Claude writes the fix, CI blocks the merge until it's resolved.
- Strix
- Claude
- GitHub Actions
- SAST
- A pull request triggers the GitHub Actions workflow
- Strix scans the code and validates real exploits
- Claude writes a fix for each confirmed finding
- The merge stays blocked until the finding is resolved
03
Road Semantics
Python pipeline that turns vehicle speed and GPS logs into RDF road-quality data and plain-English reports.
20 km field-tested
- Python
- Raspberry Pi
- RDF (rdflib)
- SPARQL
- Raspberry Pi watches the sensor CSV and classifies each reading
- Turtle file reaches the aggregation point and loads with the ontology
- R-tree matches points to known locations and averages speeds
- Report generator runs SPARQL and reverse-geocodes with Geoapify
- Text report is uploaded to the Flask API on the web server
04
Product Knowledge RAG Bot
Production RAG chatbot that answers user questions from help articles and 1 GB of tabular data.
~1 GB of data indexed in production
- Python
- LangChain
- OpenAI GPT-4 Turbo
- OpenAI Embeddings
- Ingestion embeds articles and CSV rows into Pinecone
- User sends a question to the RAG bot
- Question is embedded, top 3 chunks retrieved from Pinecone
- GPT-4 Turbo answers from the retrieved chunks only
Experience
Where I've worked
Pragnakalp Techlabs(opens in a new tab)
Jan 2024 - PresentBackend AI Engineer
Jul 2024 - Present · Gujarat, India
- Design and deploy GenAI applications (chatbots, RAG systems, voice agents) with OpenAI, Anthropic Claude, and Google Gemini.
- Build and host backends on AWS (EC2, RDS, Lambda), GCP, and Azure.
- Migrated an AI analytics SaaS from MySQL to PostgreSQL and overhauled query patterns, cutting the slowest endpoint by 70-80%.
Software Development Intern
Jan 2024 - Jun 2024 · Gujarat, India
- Built RAG chatbots and an OpenAI Assistants marketing app.
- Migrated CRM data from Copper CRM to GoHighLevel with a Python pipeline.
Skills
What I work with
AI and LLMs
- OpenAI
- Anthropic Claude
- Google Gemini
- LangChain
- RAG
- Prompt engineering
- Voice AI agents
- YOLO
- CLIP
- OCR
Backend and Cloud
- Python
- Flask
- FastAPI
- Django
- Celery
- Redis
- Docker
- AWS
- GCP
- Azure
- Nginx
- uWSGI
Data and Vector Stores
- PostgreSQL
- MySQL
- MongoDB
- Elasticsearch
- pgvector
- Qdrant
- Milvus
- Pinecone
- ChromaDB
Security and Automation
- Strix
- AST-based SAST
- GitHub Actions
- n8n
- Zapier
- Selenium
- Claude Code
- Cursor
Certifications
Credentials
Verify Claude Certified Architect, Foundation (opens in a new tab)Claude Certified Architect, Foundation
Courses
Introduction to Model Context Protocol
Verify Introduction to Model Context Protocol (opens in a new tab)Claude Code in Action
Verify Claude Code in Action (opens in a new tab)Building with the Claude API
Verify Building with the Claude API (opens in a new tab)AI Fluency: Framework & Foundations
Verify AI Fluency: Framework & Foundations (opens in a new tab)
Blog
Latest writing
· 1 min read
Cutting code review noise with Jev, a millisecond decision model
Adding Jev, a non-autoregressive decision model, as the final gate in a 3-tier automated PR review pipeline to filter out low-signal findings.
· 1 min read
How I reach my homelab database from anywhere, without exposing it to the internet
How I reach Postgres on my homelab from anywhere with Tailscale and PgBouncer, without opening a single port to the internet.
· 1 min read
Your homelab doesn't need port forwarding. It needs Tailscale.
Tailscale gives a homelab private, encrypted remote access without a public IP or port forwarding, and it's a low-risk way to learn real networking.
Let's work together
I'm open to backend and AI engineering roles. The fastest way to reach me is email.