KLS · 2026

SOFTWARE ENGINEER · BUILDING PRODUCTS

I build production
software that solves
real problems.

Software engineer working across applied AI, backend systems, and product. I take problems from first sketch to something running in production — measurable, tested, and used.

Or get in touch

Based

Bhubaneswar, IN

Graduating

2026 · KIIT

Focus

AI · Full-Stack

Status

Open to roles

LIVE · FROM HIRESENSE AI

HIRESENSE · LIVE

VIEW LIVE DEMO

senior ML engineer with RAG experience

⌘K

RANKED · RRF(FAISS ⊕ BM25)SEARCHING…

Priya S.ML Engineer · 4y

RAGFAISSPyTorch

0.94

Marcus T.Full-Stack · 6y

Next.jsNodeAWS

0.88

Ana L.AI Engineer · 3y

LLMsVectorFastAPI

0.86

Rahul D.Backend · 5y

PostgrespgvectorGo

0.79

Sofia M.ML Engineer · 2y

NLPKeras

0.71

Interactive preview · not a mockupSee the project →

SCROLL

LLMsRAGVector SearchFAISSNext.jsFastAPITensorFlowPyTorchSupabasepgvectorVercelHugging FaceTypeScriptSystem DesignLLMsRAGVector SearchFAISSNext.jsFastAPITensorFlowPyTorchSupabasepgvectorVercelHugging FaceTypeScriptSystem Design

01ORIGIN

From first line of code to production.

Five years of building — from foundations in CS to shipping production AI. The path, in five moments.

KLS · 2026

“Build things that work. Ship them. Learn what breaks. Repeat.”

— WORKING PRINCIPLE

2022
Started B.Tech, Computer Science
Joined KIIT University, Bhubaneswar. Built a strong foundation in programming, algorithms, and core computer science.
2023
AI and full-stack, side by side
Explored AI and full-stack development through practical projects using Python, machine learning, and modern web technologies.
2024
End-to-end software
Shipped end-to-end projects across backend systems, APIs, deployment, and machine learning — building the muscle for real products.
2025
Applied AI and RAG
Focused on applied AI, multimodal machine learning, and retrieval-augmented generation through academic and personal projects.
2026
Shipping production AI
Built and deployed production-ready software including HireSense AI and other AI-powered applications. Open to Software Engineer and AI Engineer roles.

02SELECTED WORK

Products, not demos.

Four projects, shipped end-to-end. Each one solves a real problem — not a resume line.

01FLAGSHIP · LIVE

Live

HireSense AI. Grounded.

Hybrid retrieval for grounded hiring decisions — shipped end-to-end.

FAISS vector search fused with lexical retrieval via Reciprocal Rank Fusion, wrapped in a stateless FastAPI service and a Next.js 15 UI. Fully live, publicly used.

377

records indexed

<1s

query latency

Live

in production

Next.js 15React 19FastAPIFAISSRRFTypeScriptVercelHugging Face

HIRESENSE · LIVE

VIEW LIVE DEMO

senior ML engineer with RAG experience

⌘K

RANKED · RRF(FAISS ⊕ BM25)SEARCHING…

Priya S.ML Engineer · 4y

RAGFAISSPyTorch

0.94

Marcus T.Full-Stack · 6y

Next.jsNodeAWS

0.88

Ana L.AI Engineer · 3y

LLMsVectorFastAPI

0.86

Rahul D.Backend · 5y

PostgrespgvectorGo

0.79

Sofia M.ML Engineer · 2y

NLPKeras

0.71

02PLATFORM

service_agreement_v4.pdfPage 3 of 47

The Parties agree that confidential information shall be held for five (5) years following termination.

Payment terms are Net 30 from invoice date, subject to 1.5% monthly late fee.

Liability capped at fees paid in the preceding twelve months...

Governing law of the State of Delaware, without regard to conflict...

3 clauses extracted · 42ms

Legal AI Platform. Cited.

Semantic search and clause extraction across 1,000+ legal documents.

Next.js + Supabase (pgvector) RAG pipeline. LLM-powered clause extraction. Signed URLs, edge APIs, SWR frontend for real-time analysis.

1,000+

documents

~50ms

retrieval

<1s

query

Next.jsSupabasepgvectorAuth0LLM APIsSWREdge Runtime

03RESEARCH

Neural Digital Twin. Multimodal.

Multimodal EEG + MRI diagnostic AI for early cognitive decline.

CMF-ViT on EEG (TUH dataset) and EfficientNet-B4 on MRI (ADNI). Unified diagnostic decision layer.

PyTorchVision TransformerEfficientNet-B4EEGMRITUHADNI

EEG · CMF-VIT · TUH

88–92%

accuracy · 22 patient records

MRI · EFFICIENTNET-B4 · ADNI

97.6%

test acc. · 111 records

PROBLEM

Early detection of cognitive disorders requires fusing signals across modalities — rarely done well.

04TIME-SERIES

LSTM Forecasting. Stable.

3-layer LSTM forecasting with rigorous validation and stability tuning.

3-layer LSTM on 5,000+ points with EarlyStopping and LR scheduling. Tuned for stability, not just accuracy.

0.96

R²

66.5

RMSE

−70%

val loss

TensorFlowKerasNumPyPandasMatplotlib

FORECAST vs ACTUAL · 60 STEPS actual predicted

R²

0.96

RMSE

66.5

MAE

31.06

03CRAFT

The tools I reach for when shipping.

Not a laundry list. The stack I actually use, grouped by how I think about a problem.

AI / ML

01 / 06

·LLM APIs·RAG Pipelines·Vector Embeddings·FAISS·Reciprocal Rank Fusion·CMF-ViT·EfficientNet·LSTM·CNN·TensorFlow·Keras·scikit-learn

Backend

02 / 06

·Python·FastAPI·Node.js·Express·Java·C++·REST·System Design

Frontend

03 / 06

·Next.js 15·React 19·TypeScript·TailwindCSS·Framer Motion·SWR

Data

04 / 06

·PostgreSQL·pgvector·Supabase·MySQL·SQL

Cloud / DevOps

05 / 06

·Docker·Vercel·Railway·Hugging Face·CI/CD·Git

Engineering

06 / 06

·Pytest·MyPy·ESLint·DSA·OOP·Agile

Production Projects

Models Trained

Technologies

GitHub Commits

0yrs

Engineering

04 — CONTACT

Let’s build something worth building.

I read every message. Fastest way to reach me is email — the terminal on the right works too, if you prefer keyboards.

kanhaiyals2019@gmail.com

+91 9123678922

GitHub LinkedIn Download Resume PDF

kanhaiya — terminalready

Welcome. Type "help" for commands, or "hire" to reach me directly.

I build productionsoftware that solvesreal problems.