Prathamesh Gokulkar

Hi, I'm Prathamesh

AI Engineer building production-grade AI applications, LLM pipelines, RAG systems, and agentic workflows.

About

I'm an AI Engineer focused on building practical AI systems using LLMs, RAG, agentic workflows, and modern backend infrastructure.

I've worked across document intelligence, healthcare chatbots, LLM pipelines, model fine-tuning, and production AI applications.

My projects include SmartScan AI, a multi-agent document intelligence system that achieved 98% OCR accuracy and ~5s latency, and NutriGuide, where I optimized a RAG pipeline by tracing and eliminating database bottlenecks.

I enjoy working at the intersection of AI, software engineering, and product requirements — taking an idea from a problem statement to a working system.

Currently, I'm an AI Engineer at Intlnc AI, building AI applications against real client requirements.

Highlights

3rd / 48

Technokratia 2026

98% OCR

SmartScan AI accuracy

~5 sec

SmartScan AI latency

40%

Less document parsing effort

75%

VRAM reduction, code assistant

~2 ms

PostgreSQL execution after optimization

Intlnc AI

AI Engineer · Present

Production experience

Currently building AI applications at Intlnc AI. Each shipped system will be added here with the problem, architecture, engineering decisions, and impact — only after it has actually shipped.

Intlnc AI

AI Engineer · Pune

Sep 2026 – Present

Building production AI applications from client requirements through technical breakdown, development, and delivery.

Working on production AI applications with high ownership
Taking work from client requirements to a technical breakdown, implementation, and application delivery
Exposure to enterprise and client-facing AI workflows
Joining discussions with senior stakeholders and turning business requirements into engineering work
Explaining technical AI concepts in language non-technical stakeholders can use
Working inside real delivery constraints that don't show up in personal projects

Experience

Qualitia Software

AI / Software Intern · Remote

Feb 2026 – Apr 2026

Worked on LLM pipelines where the job was control over the model, the cost, and the architecture, not only a call to a proprietary API.

Worked on LLM pipelines and locator automation
Migrated from proprietary APIs toward self-hosted open-weight 7B models
Worked through LLM inference and model deployment considerations
Helped move the system toward a decoupled architecture
Worked on reducing dependency on external and proprietary model APIs

SigmoidAI

AI / ML Intern · Remote

Nov 2025 – Jan 2026

Early production exposure: a hospital chatbot and dashboard that had to run as a deployed service.

Built a WhatsApp-based hospital management chatbot
Deployed services on GCP Cloud Run
Built and connected backend services for the product
Worked on the hospital dashboard the chatbot fed into
Got early exposure to deploying an AI application in a real environment

What I build

I build AI systems that…

Understand unstructured data

Document intelligence, OCR, extraction, and validation.

Retrieve and reason over knowledge

RAG pipelines using embeddings, vector search, and structured databases.

Use multiple specialized agents

Agentic workflows with orchestration, validation, and task-specific agents.

Run efficiently

Model optimization, quantization, LoRA/PEFT, database optimization, and latency reduction.

Move beyond prototypes

FastAPI backends, Dockerized applications, cloud deployment, and production-oriented architecture.

Skills

AI / LLM

LLM Applications

RAG

Multi-Agent Systems

LangChain

LangGraph

Prompt Engineering

LLM Evaluation

Embeddings

Vector Search

LoRA / PEFT

Quantization

Hugging Face

Backend

Python

FastAPI

REST APIs

PostgreSQL

MongoDB

AI Infrastructure

Docker

GCP Cloud Run

AWS

Vector Databases

Model / API integration

AI application deployment

Databases / Retrieval

PostgreSQL

ChromaDB

FAISS

Pinecone

Sentence Transformers

Frontend

React

Next.js

Observability

LangSmith

Git / GitHub

Selected work

Systems with measured results

Document intelligence, a RAG pipeline I traced and fixed, and a fine-tuned code assistant. Architecture and numbers, not a demo list.

Source

SmartScan AI

3rd place · 48 teams

Technokratia 2026

Designed an agentic document-processing pipeline where specialized agents handle OCR, extraction, reasoning, and validation instead of a single monolithic LLM call.

Document → Orchestrator → OCR → Extraction → LLM agent → Validation → Structured output

98%

OCR accuracy

~5 sec

Latency

40%

Less parsing effort

3rd / 48

Technokratia 2026

Python

LangGraph

LangChain

OCR

RAG

ChromaDB

FastAPI

Demo

Source

NutriGuide

Traced the bottleneck, then fixed it

RAG performance case study

Personalized nutrition assistant with onboarding, TDEE and macro recommendations, and recipe retrieval through a RAG pipeline over embeddings, ChromaDB, and PostgreSQL.

Instrumented the pipeline with LangSmith and traced a ~26s response to a sequential ILIKE scan across 231K+ recipes. A trigram GIN index cut database execution to ~2ms and application-level lookup to ~9–30ms after warm-up.

Next.js → FastAPI → RAG → Embeddings → ChromaDB → PostgreSQL

~26s

Response before tracing

~2 ms

Query execution after

9–30 ms

Lookup after warm-up

231K+

Recipes in search

Next.js

FastAPI

LangChain

LangSmith

ChromaDB

PostgreSQL

Demo

Source

DevTalk

75% VRAM reduction

Fine-tuned code assistant

Fine-tuned a 600M-parameter Qwen model with LoRA/PEFT and quantization on a T4 GPU, then built a coding assistant around that model. Parameter-efficient fine-tuning and quantization cut VRAM use by about 75%. The broader project also included speech-related functionality.

75%

VRAM reduction

600M

Qwen parameters

LoRA

PEFT fine-tune

Training GPU

Python

Hugging Face

PyTorch

LoRA / PEFT

Quantization

Contact

Get in Touch

Want to chat? Just shoot me a dm with a direct question on twitter and I'll respond whenever I can.