Meridian
activeOffline-first productivity platform with AI voice capture, two-way Google Calendar and Outlook sync, and real-time sync across web and mobile.
FastAPI / React Native / Expo / Supabase / PostgreSQL / Redis / Whisper API
Narayan Joshi
Full-Stack & AI Engineer
I build the backend under AI products — RAG pipelines, LLM orchestration, and APIs that hold up in production.
About
Final-year B.E. student at Sinhgad Institute of Technology & Science, Pune. I work where AI logic meets backend infrastructure: RAG systems, LLM orchestration, and scalable API design built on Django and vector databases.
Right now I'm building Meridian, a productivity app with AI-powered voice transcription, and DocuMind, a production-grade RAG platform with hybrid search and evaluation pipelines.
backend
Python / Django / DRF / FastAPI / Node.js / JWT auth / Celery / Redis / PostgreSQL / Supabase / WebSockets
ai & ml
RAG / LLMs / OpenAI / Gemini / Llama / Whisper API / semantic search / pgvector / FAISS / AI agents / DeepEval
frontend & mobile
React / React Native / Expo
tools & devops
Docker / Git / Linux / Postman / Power BI / FCM
Projects
DocuMind
activeMulti-tenant RAG platform: FAISS vector search with cross-encoder reranking, bring-your-own-key LLM routing, and DeepEval generation scoring.
Django / React / FAISS / PostgreSQL / DeepEval
ARGUS
activeFastAPI backend that fans each query out to researcher, analyzer, and verifier roles, then synthesizes their structured output into one answer.
Python / FastAPI / LLM routing / AI agents
Research Radar
Paper search engine for computer vision and LLM research: ParadeDB BM25, RRF hybrid ranking, and semantic find-similar, covered by 105 tests.
FastAPI / Next.js / PostgreSQL / ParadeDB / pgvector
WindLens
Decision support for day-ahead wind power buying on the GB grid: it pairs forecasts with reported outturn and translates the misses into plain language.
TypeScript / Elexon data / Vercel
AirLock DB
Local-first migration CLI for air-gapped environments: strict schema validation, SQLite or JSONL output, and one self-contained HTML report.
Python / SQLite / JSONL
Open source
Pull requests and bug reports in projects I don't maintain.
open-telemetry/opentelemetry-collector
- merged Reuse cached item counts for trace and log max-size splits — the batch processor stops re-counting every item on max-size splits
- merged Reuse cached metric count for max-size splits — same fix for the metrics pipeline
- open Fail startup when persistent queue metadata cannot be read
- open Fix persistent queue state after marshal errors
- open Document that ProfilesCountSizer reports two different units
- plus a series of filed issues on persistent-queue and batching correctness
pingdotgg/t3code
- open perf(client-runtime): reuse projection entity lookups
- open fix(opencode): recover model catalogs during plugin activation
- plus four more PRs closed after review: #20328, #18304, #15863, #13606
Experience
Jan — Mar 2025
Power BI Intern
KasNet Technologies Pvt. Ltd. — Pune
Designed interactive dashboards and reports, turning complex datasets into clear visual narratives for business decisions.
May — Aug 2024
Data Science & Analytics Intern
Zidio Development Pvt. Ltd. — Remote
End-to-end analysis in Python and SQL. Automated data cleaning workflows, cutting manual preparation time by 40%, and built predictive models and visualizations for stakeholders.
Education
2022 — 2026
B.E. in Electronics & Telecommunication
Sinhgad Institute of Technology & Science, Narhe, Pune
T.E. CGPA: 8.55
2022
HSC (12th grade)
S B Junior College, Jalna
58.83%
Certifications
IBM
Enterprise Grade AI
Deloitte Australia
Data Analytics Job Simulation
Spoken Tutorial
C / C++ Programming
Contact
I'm looking for Full-Stack, Backend, AI Engineer, and Gen AI roles. If you have a question, a project, or an opening — my inbox is open.