About Me
I'm Kishna Kushwaha — an AI/ML engineer and data analyst based in Noida, India. Before building AI systems, I spent 7 years at PolicyBazaar analyzing large-scale financial, conversion, and operations data. That experience shaped how I build today: obsessed with real-world outcomes and deployable AI systems instead of notebook-only experiments.
🚀 What I Build
- Production-Ready AI Systems — Designing and deploying RAG pipelines, embedding-based semantic retrieval architectures, and multi-LLM orchestration engines to real-world servers (AWS EC2, Render).
- Advanced LLM Pipelines — Engineering complex fact-checking systems, fake news classifiers, and semantic similarity engines with structured JSON outputs.
- Scalable Backend & Desktop Apps — Building robust FastAPI backends with SSE streaming, paired with seamless frontend interfaces and Electron.js desktop clients.
- Data Intelligence & Analytics — Creating dynamic KPI dashboards, automated clustering, and customer segmentation tools grounded in deep financial operations experience.
🛠 Tech Stack
💼 Experience
Data Analysis & Operations
Analyzed large-scale customer conversion and financial datasets to guide business strategy. Built reporting pipelines used by senior leadership and led a team of 15 people to manage coordination between sales and analytics.
💻 Live Projects
Content Refactoring Engine (CRE SaaS)
- • Designed semantic similarity pipeline using sentence embeddings and cosine similarity matrices
- • Implemented triple-redundant multi-LLM inference architecture (DeepSeek → Groq → OpenRouter)
- • Built scalable FastAPI backend with SSE streaming interface
- • Developed composite scoring engine combining statistical similarity metrics with LLM semantic reasoning
- • Implemented structured HTML-to-JSON transformation pipeline
- • Integrated authentication, API usage tracking, and persistence layer
Tech Stack: Python, FastAPI,
Sentence-Transformers, NumPy, Scikit-learn, SQLite, DeepSeek API, Groq API, OpenRouter API,
Gemini API, SSE Streaming UI, Render
Skills Used: Semantic Similarity,
Embedding Pipelines, Cosine Similarity Matrices, Multi-LLM Routing, Prompt Engineering,
Retrieval Pipelines, HTML-to-JSON Transformation, API Authentication, Usage Tracking Systems
IntervuAI — Context-Aware Interview Assistant
- • Built retrieval-augmented assistant supporting contextual technical responses
- • Developed FastAPI backend supporting Groq Llama-3 inference
- • Implemented embedding-based semantic retrieval architecture
- • Designed Electron desktop client with screen-sharing capability
- • Created modular inference + embeddings + speech-to-text service architecture
Tech Stack: Python, FastAPI, Electron.js,
Sentence-Transformers, Groq API, Render
Skills Used: RAG Pipelines, Semantic
Retrieval, Session Memory Architecture, Context Window Engineering, Desktop AI App
Development, LLM API Integration, Speech-to-Text Integration
Customer Segmentation Dashboard
- • Performed RFM feature engineering and K-Means clustering segmentation
- • Built automated segment classification engine
- • Designed interactive KPI dashboards using Plotly
Tech Stack: Python, Pandas, NumPy,
Scikit-learn (KMeans, StandardScaler), Streamlit, Plotly
Skills Used: Customer Segmentation, RFM
Analysis, Feature Engineering, Clustering Algorithms, Predictive Segment Classification, KPI
Dashboarding
VeriFact — AI Fact-Checking Pipeline
- • Built multi-stage claim verification pipeline including stance detection workflow
- • Implemented SBERT semantic retrieval with BART-MNLI reasoning engine
- • Designed credibility-weighted evidence scoring architecture
- • Dockerized deployment using GitHub Actions CI/CD to AWS EC2
- • Developed pytest-based unit and integration testing suite
Tech Stack: Python, Flask, HuggingFace
Transformers, SBERT, BART-MNLI, Tavily API, DuckDuckGo API, Docker, GitHub Actions, AWS
EC2
Skills Used: Evidence Retrieval Pipelines,
Stance Detection (NLI), Semantic Search, Multi-Source Evidence Aggregation, CI/CD
Automation, Containerization, Model Evaluation Pipelines
Fake News Detection LLM System
- • Built semantic retrieval + stance-aware classification pipeline
- • Integrated SBERT embeddings with BART-MNLI reasoning workflow
- • Implemented structured inference pipeline supporting real-time classification
- • Deployed scalable inference API using Render
- • Optimized fallback prompt orchestration across multiple LLM providers
Tech Stack: Python, FastAPI, SBERT,
BART-MNLI, Render
Skills Used: Semantic Retrieval Pipelines,
Stance Detection, Evidence-Based Classification, Real-Time Inference APIs, Prompt
Optimization, Multi-Model Routing Logic