Hi there, I'm

Heet Dobariya

I'm an engineer who builds production AI.

Right now I'm a founding engineer at Serin Labs, working on Python backends, LLM pipelines and RAG over vector and graph databases.

Ask AI about me

View my work Resume Get in touch

Ahmedabad, India · --:-- IST

Scroll

1About

Where backend engineering meets applied AI.

I'm a Computer Science graduate from Pandit Deendayal Energy University (Minor in IoT) and a founding engineer at Serin Labs, where I help build Serin AI from the ground up.

Most of my work sits where backend engineering meets applied AI: FastAPI services, retrieval pipelines over Pinecone, Qdrant and Neo4j, and the evaluation work that keeps LLM output honest.

2Experience

The road so far.

From internships to founding-team work on production AI.

Mile 04 · Now
Founding Engineer · Serin Labs
Sep 2026 – Present · Ahmedabad
- Building core Serin AI features as part of the founding team, shipping fixes and platform improvements.
- Developed Serin MCP integrations and led platform-wide QC to catch bugs, edge cases and usability issues.
- Serin AI
- MCP
- QC
Mile 03
Jr. Software Engineer · Growexx AI Solutions
Jan 2026 – Jul 2026 · Ahmedabad
- Developed Python/FastAPI backend features for Ojavix, a RAG-based AI therapist chatbot.
- Built the knowledge base from PDF/TXT files via Azure Document Intelligence, indexed in Pinecone and Neo4j.
- Developed therapy workflows combining vector/graph retrieval with short- and long-term user memory.
- FastAPI
- RAG
- Pinecone
- Neo4j
- Azure
Mile 02
Data Science Intern · Blink Analytics
Feb 2025 – Jan 2026 · Ahmedabad
- Curated domain-specific datasets to fine-tune LLMs with SFT and RLHF, improving accuracy and relevance.
- Evaluated and debugged LLM outputs in production using custom benchmarks and interpretability tools.
- LLMs
- SFT
- RLHF
- Evals
Mile 01
AI/ML Intern · Nexus Info
May 2024 – Jun 2024 · Remote
- Built TravelGuru and EduAdvisor chatbots for travel guidance and college admissions, and an ML-based disease prediction system.
- Gemini API
- Streamlit
- scikit-learn

3Projects

Selected builds.

Production AI, RAG systems, NLP tools and a little hardware.

HD-SCAN · ProjectsModules 5/5 · OK

Projects Serin AIWork · Serin Labs

Serin AI

AI interview platform

An AI-powered interview platform that runs voice and video interviews, proctors them for integrity and evaluates every candidate, so hirers can interview everyone and shortlist the best. I'm on the founding team, working across the interview agent, reports, the MCP server and the web app.

Live data

Role: Founding Engineer
Interviews: Voice · Video
Access: Web · MCP

Modules

Next.js
React
FastAPI
LLM agents
MCP
Supabase
PostHog

Sequence

01Job & candidatesRUN
02AI interview--
03Proctoring--
04Evaluation--
05PDF report--
06Hirer dashboard--

My work

Extended the Serin MCP server's tools and skills for the new assessment report
Built report features: additional-assessment section, PDF timeline monitoring and anomaly rows
Tuned the interview agent: prompt rules, turn intake and transcript generation
Hardened the interview flow: start-failure handling and the Report Issue form
Set up Infisical secrets, PostHog analytics and keyboard shortcuts on the hirer app

Projects QSkim

QSkim

Govt. scheme discovery platform

Full-stack platform to discover and compare Indian government schemes, using hybrid RAG with BGE-M3 and Qdrant, metadata-based eligibility matching, and multilingual voice and WhatsApp access.

Live data

Retrieval: Hybrid
Access: Web · Voice · WhatsApp
Matching: Eligibility-aware

Modules

RAG
BGE-M3
Qdrant
FastAPI
Next.js
Groq
Sarvam
Twilio

Sequence

01Scheme dataRUN
02BGE-M3 embeddings--
03Qdrant hybrid search--
04Eligibility filter--
05Groq answer--
06Voice / WhatsApp--

Highlights

Profile-based scheme recommendations and grounded Q&A over Indian government schemes
Hybrid retrieval: BGE-M3 embeddings in Qdrant with metadata filters for eligibility
Knowledge base built by scraping scheme portals with Playwright and BeautifulSoup
Multilingual voice via Sarvam and WhatsApp access via Twilio

Projects IoT Fire Detection

IoT Fire Detection

On-device TinyML on an ESP32-CAM

Low-cost fire detection running a TinyML model on an ESP32-CAM, trained with Edge Impulse on a custom 176-image dataset. On detection it sounds a buzzer and switches a relay-driven water pump, while DHT11 and MQ-2 monitor heat and gas.

Live data

Dataset: 176 images
Model input: 96 × 96
Trigger: > 0.8 confidence

Modules

ESP32-CAM
Edge Impulse
TinyML
DHT11
MQ-2

Sequence

01Camera frameRUN
02TinyML model--
03Fire? > 0.8--
04Buzzer--
05Relay--
06Water pump--

Highlights

Collected 176 training images with the ESP32-CAM itself (93 fire, 83 no-fire)
96×96 TinyML model trained in Edge Impulse, running fully on the board
Fire confidence above 0.8 sounds a buzzer and switches a relay-driven 9V pump
DHT11 and MQ-2 monitor temperature, humidity and gas alongside the camera

Projects TalentLens

TalentLens

Resume ranking system

Ranks resumes against job descriptions using SpaCy skill extraction and BERT embeddings, scoring relevancy with Gradient Boosting, Random Forest, Ridge and SVR models on a Flask + MongoDB backend.

Live data

Embeddings: BERT
Models: 4 regressors
Output: Ranked shortlist

Modules

BERT
SpaCy
scikit-learn
Flask
MongoDB

Sequence

01Resume + JDRUN
02SpaCy parsing--
03Skill extraction--
04BERT similarity--
05Regressor ensemble--
06Ranked list--

Highlights

Parses resumes and extracts skills with SpaCy
Scores semantic fit to the job description with BERT sentence embeddings
Combines Gradient Boosting, Random Forest, Ridge and SVR for relevancy scoring
Flask + MongoDB backend for storing and ranking candidates

Projects Scholaris

Scholaris

Assignment evaluator

Flask + MongoDB app where teachers create and grade assignments. The Groq API generates viva questions from each submitted file (TXT, PDF, DOCX) so teachers can assess students.

Live data

Formats: TXT · PDF · DOCX
LLM: Groq API
Output: Viva questions

Modules

Flask
MongoDB
Groq API

Sequence

01AssignmentRUN
02Student upload--
03Document parsing--
04Groq LLM--
05Viva questions--
06Teacher review--

Highlights

Teachers create assignments; students submit TXT, PDF or DOCX files
Submissions are parsed and sent to the Groq API
Generates viva questions from each submission so teachers can assess students
Flask + MongoDB backend with a Bootstrap and Jinja2 interface

↑ ↓ Select

Visit site

4Skills

What's in the garage.

Languages

Python
C
C++
SQL
Java
JavaScript
TypeScript
HTML
CSS

AI / ML

LLMs
RAG
Prompt Engineering
NLP
Transformers
BERT
Deep Learning
scikit-learn
TensorFlow
Keras

Backend & Data

FastAPI
Flask
PostgreSQL
MongoDB
Pinecone
Qdrant
Neo4j

Cloud & Tools

Git
GitHub
AWS
Azure
Next.js

IoT

ESP32
Arduino
Edge Impulse

Certifications

GATE: Data Science & AI · AIR 3876 · Score 4322025
Neural Networks Bootcamp · Udemy2026
SQL & PostgreSQL · Udemy2026
Foundations of Agents · Hugging Face2025
Machine Learning Specialization courses · Stanford Online2024

Leadership

Content & Documentation Head

Mind Ripple, PDEU

May 2024 – Jun 2025

NContact

Let's take it for a spin.

Open to AI engineering roles, interesting projects and good conversations.

Elsewhere

Heet Dobariya

Founding Engineer · Serin Labs

Jr. Software Engineer · Growexx AI Solutions

Data Science Intern · Blink Analytics

AI/ML Intern · Nexus Info

Serin AI

Live data

Modules

Sequence

My work

Languages

AI / ML

Backend & Data

Cloud & Tools

IoT

Certifications

Leadership