Skip to content
SK

About

From a data science degree to building agentic systems in production.

AI/ML Engineer with 1+ years of combined full-time and internship experience building enterprise Generative AI, agentic AI, and RAG systems for pharmaceutical and business intelligence use cases. Experienced in LangGraph, LangChain, CrewAI, MCP, Graph RAG, LLM evaluation, and FastAPI. Contributed to enterprise AI platforms involving clinical knowledge retrieval, multi-agent data structuring, safety guardrails, and natural-language-to-SQL workflows. Strong foundation in Python, machine learning, NLP, and LLM application engineering.

Career journey

How I got here

Three stages, and the thing each one taught me.

2021 — 2024

B.Tech, Computer Science (Data Science)

TKR College of Engineering and Technology

Four years on the data science specialization - the statistics, machine learning and deep learning foundation that the later LLM work sits on top of. Alongside the coursework: NPTEL certifications from IIT Madras and IIT Guwahati, and a first-place finish at the AIML EduNext Hackathon.

Feb 2025 — Apr 2025

AI Engineer Intern

BiHub Solutions (Innovative Office Solutions)

First production LLM work. The problem was dashboard migration - hundreds of Tableau workbooks that needed rebuilding in Power BI by hand. The answer turned out to be a split one: deterministic XML parsers for structure, LLM agents only for the part that genuinely needs judgment, expression translation. It reached 85-90% conversion accuracy under the project's evaluation criteria, and taught me where an LLM belongs in a pipeline and where it does not.

Sep 2025 — Present

ML Associate — AI/ML Engineer

Avira Digital Technologies

Enterprise agentic AI, at the scale where the hard problems stop being about prompts. Contributing across an internal platform - a provider-agnostic LLM gateway, MCP tool servers, layered agent memory, safety guardrails and an LLM-as-judge evaluation harness - and to client retrieval systems in the pharmaceutical domain, including a clinical knowledge graph RAG and a multi-agent data structuring platform.

Current focus

What I am thinking about right now

Retrieval past the naive baseline

Most RAG demos work because the question and the answer share vocabulary. The interesting cases are the ones where they do not - relationship questions, multi-hop reasoning, and domains where the structure of the data matters more than the wording of any single chunk.

Agents that can be measured

Agent quality is easy to assert and hard to demonstrate. Judge harnesses, labeled datasets and shadow evaluation are what turn “this feels better” into something you can regress against between iterations.

Cost and latency as design constraints

Routing every query to the largest model is a decision, not a default - and usually the wrong one. Complexity classification, cost-tiered dispatch and caching are where a system stops being a demo.

Technical background

The full stack I work across

Grouped the way I actually use them, rather than as one undifferentiated list.

Languages

2

Primary day-to-day tooling.

PythonSQL

Generative AI

10

Agent orchestration, retrieval, and evaluation.

LangGraphLangChainCrewAIMCPMulti-Agent OrchestrationRAGGraph RAGLLM EvaluationGuardrailsPrompt Engineering

Machine Learning

9

Model training, fine-tuning, and classical ML.

Machine LearningDeep LearningNLPComputer VisionLLM Fine-TuningPyTorchTransformersScikit-learnTensorFlow

Data & Retrieval

11

Graph and vector stores behind the retrieval layers.

Neo4jAWS NeptuneCypherGremlinPostgreSQLMySQLQdrantFAISSPineconeBM25Hybrid Search

Backend & Cloud

7

Service layers, deployment, and experiment tracking.

FastAPIDockerAWS (EC2, S3, SageMaker, Bedrock)GitRedisMLflowOptuna

Data & BI

5

Analysis, dashboards, and internal tooling.

NumPyPandasMatplotlibPower BIStreamlit

Education

Academic background

2021 — 2024

B.Tech, Computer Science

Data Science Specialization

TKR College of Engineering and Technology

Hyderabad, India

Certifications & achievements

Credentials

Data Science for Engineering

NPTEL (IIT Madras)

ML & Deep Learning Fundamentals

NPTEL (IIT Guwahati)

1st Rank — AIML EduNext Hackathon

AIML EduNext

Achievement

Professional goals

Where I am pointed next

  • Go deeper on agent reliability - planning, memory and recovery when a tool call fails.
  • Keep building evaluation infrastructure, because it is the part that compounds.
  • Work on retrieval systems where the domain structure is genuinely hard, not just large.
  • Take on selected freelance builds that push into territory my day job does not cover.

Curious about the rest?

The AI Lab page covers what I am reading, experimenting with and learning outside of shipped work — including the areas I am still exploring rather than claiming.

Visit the AI Lab