Abhisek Sahoo
Founding-Engineer

AS
Available to Work

About

Final Year CS undergrad with experience interning at startups and taking products from concept to launch. Proficient in the
TypeScript ecosystem
and modern product engineering, with a focus on GenAI using the latest models on
FalFal
and
HuggingFaceHugging Face

Work Experience

E

Emergex Labs

June 2025 - Sep 2025
Founding Engineer
C

Colestore

July 2024 - Aug 2024
Software Engineering Intern
S

Stealth Startup

May 2024 - June 2024
Software Engineering Intern

Education

B

Bennett University

2022 - 2026
Btech in Computer Science and Artificial Intelligence

Primary Stack

Latest Projects

JIF.AI

JIF.AI

AI Content Engine
2024

AI-native media OS where autonomous agents clip, repurpose, and publish viral content at scale

Next.js
Typescript
AWS
FastAPI
Docker
Groq
Vercel AI SDK
GCP
PostgreSQL
DrizzleORM
ExaAI
JIF.AI
Thriftwise

Thriftwise

Agentic MCP Chat
2025

GenAI e-commerce web search engine

Next.js
Vercel AI SDK
Typescript
PineConeDB
PostgreSQL
DrizzleORM
TailwindCSS
ExaAI
Tanstack Query
AWS S3
OpenAI SDK
Stripe
Thriftwise
Infratherm

Infratherm

IRT Crack Detector
2024

IRT based pavement crack detection system for road maintenance

Next.js
Typescript
FastAPI
Docker
TailwindCSS
Shadcn UI
Tensorflow
Render
OpenCV
Infratherm
Devconnect

Devconnect

Social Platform for Indie Devs
2024

Connect, Collaborate and Code with developers for hackathons and projects

React Native
Expo
Appwrite
NativeWind
Devconnect

Research Work

Infrared Thermal Imaging for Pavement Inspection with Heirarchical Hybrid Vision Transformers

This paper benchmarks CNN models (ResNet50, InceptionV3, EfficientNet-B0, VGG19) and ViT models (ViT-B/16), along with hybrid Heirarchical ViT-CNN architectures (CoAtNet-3, MaxViT), for IRT image based pavement crack detection. The study employs Grad-CAM for interpretability, highlighting model selection for efficient infrastructure maintenance

Enhancing Explainability in Multimodal AI via Chain-of-Thought Driven Knowledge-Infused Neuro-Symbolic Models

This paper presents an explainable multimodal AI framework for depression detection by combining Chain-of-Thought (CoT) reasoning with knowledge-infused neuro-symbolic models. The approach integrates textual, visual, and behavioral data to improve diagnostic accuracy while enhancing transparency in decision-making. By incorporating external knowledge and symbolic logic with deep learning, the model provides interpretable insights into depressive symptoms, making it more trustworthy and clinically useful.

GitHub Contributions

GitHub
Contact

Get in Touch