Deeksha Gautam

Public code / build evidence

Clean source.
Visible engineering decisions.

DG
@deekshagautam

LLM systems · Backend engineering · Applied AI

Open profile ↗

Public repositories

04 projects

Public repository

Hybrid-AI-Gateway

A policy-driven inference gateway that routes requests between local and cloud LLMs using privacy classification, task type, complexity, provider availability, response validity, and configurable rules.

View repository ↗
106Automated tests
7 / 7End-to-end evaluation
PassingGitHub Actions CI

Public repository

Local-vs-cloud-AI-lab

A practical benchmark comparing local Gemma 3 4B and cloud GPT-OSS 120B across instruction following, grounded QA, reasoning, performance, and estimated API cost.

View repository ↗
34Model requests
3Experiment groups
CompleteV1 benchmark

Public repository

Local-rag-evaluation-lab

A privacy-first local RAG system with hybrid vector and BM25 retrieval, RRF fusion, grounded Ollama generation, retrieval diagnostics, automated evaluation, regression gates, and CI.

View repository ↗
1.000Recall@4
0.893Hybrid MRR
17Evaluation cases

Public repository

Pet-thought-generator

A Streamlit GenAI application that converts structured pet context and a configurable chaos score into a dynamically constructed Gemini prompt and generated response.

View repository ↗
GeminiModel API
StreamlitApplication UI
.envSecret isolation

This portfolio

Keep the source, not just the link.

The complete GitHub Pages-ready code is packaged without dependencies, build output, or hosting identifiers.

Download source ↓