Skip to content

Tanishq Tembhurne

Hyderabad, India

tanishq.aidev1310@gmail.comLinkedInGitHub

Education

Woxsen University, Hyderabad

Computer Science Engineering (B.Tech)

2021 – 2025

Skills

Programming:
Python, JavaScript/Node.js
AI/ML:
Machine Learning, Deep Learning, PyTorch, NLP, Computer Vision, Fine-Tuning, Model Evaluation, Model Optimization, Generative AI, LLMs, SLMs
LLM & Agentic AI:
RAG, LangChain, LangGraph, LlamaIndex, AI Agents, Agentic Workflows, Prompt Engineering, Tool Calling, Multi-Agent Systems, Function Calling
AI Infrastructure & Deployment:
Azure, AWS, GCP, Azure Databricks, Docker, REST APIs, Model Serving, CI/CD, MLOps
Retrieval & Data:
Vector Databases, FAISS, ChromaDB, Neo4j, Embeddings, Semantic Search, Hybrid Search, Reranking, Pandas, NumPy
Observability & Automation:
Azure Monitor, Application Insights, Log Analytics, KQL, Azure Logic Apps, Jira Automation, Production Monitoring, Incident Response
Data Visualization:
Power BI, Matplotlib, Seaborn

Experience

Centific

SRE – Engineer

April 2025 – Present
  • Engineered an AI-driven incident investigation system across 3 Azure production environments, integrating Azure Monitor, Application Insights, Log Analytics/KQL, runbooks, and LLM-based reasoning to correlate telemetry and generate evidence-backed root-cause analyses.
  • Designed an agentic investigation workflow that dynamically analyzes alerts, queries production telemetry using KQL, retrieves relevant runbooks and historical context, and synthesizes multi-source evidence into structured RCAs, reducing MTTR by 30%.
  • Automated 30+ production alerts per month into an AI-assisted investigation pipeline using Azure Monitor, Logic Apps, and Jira, automatically initiating investigation workflows and publishing RCA findings and incident updates while reducing manual incident-management effort by 50%.
  • Built production observability and AI investigation workflows across Azure Monitor, Application Insights, and Log Analytics, combining logs, metrics, traces, dependency telemetry, and deployment activity to improve incident detection and accelerate AI-assisted diagnosis by 30%.
  • Designed an AI SRE agent with human-in-the-loop validation, generating detailed RCAs and three progressive incident updates within the 30-minute SLA, reducing L1/L2 investigation dependency by 50%.

Conquest Technology Solutions

Data Science Intern

Feb 2024 – June 2024
  • Developed and analyzed KPI dashboards in Power BI to provide actionable insights.
  • Boosted sales up to 15% in low-performing areas through effective data visualization.
  • Conducted advanced EDA to identify trends and anomalies, enhancing business intelligence.
  • Simplified complex datasets with clear visualizations and ensured accurate reporting through stakeholder collaboration.

Projects

Autonomous Meeting Agent

Python, Node.js, SLM, RAG, Chromium, OpenAI Realtime, ElevenLabs, Redis
  • Built a real-time AI virtual employee capable of autonomously joining Chromium-based meetings, actively listening to participants, answering and asking questions, following meeting instructions, sharing/presenting content, and generating and distributing meeting minutes (MoM) to attendees.
  • Fine-tuned domain-specific Small Language Models (SLMs) to power the agent’s reasoning and conversational behavior, reducing dependency on large general-purpose models while maintaining contextual response quality across meeting interactions.
  • Improved context retrieval Recall@5 from 0.52 → 0.76 by tuning chunking, embedding strategy, metadata filtering, and top-K retrieval across a 500-query evaluation set.
  • Reduced average prompt context size by 45% through conversation summarization and selective memory retrieval while maintaining 90%+ answer consistency on multi-turn evaluation scenarios.
  • Designed hierarchical short- and long-term memory for conversational context, meeting decisions, action items, and participant information; reduced average prompt context size by 45% through summarization and selective memory retrieval.
  • Implemented Redis-based caching for frequently accessed knowledge and retrieval results, reducing redundant retrieval/LLM calls and improving median response latency by 40% through cache reuse and controlled invalidation.

Legal-Aid Conversational Chatbot

LangChain, LangGraph, FAISS, Neo4j, Groq
  • Built a retrieval-augmented legal chatbot using FAISS and Neo4j, combining dense vector retrieval with knowledge-graph relationships to improve contextual retrieval across legal documents.
  • Improved Recall@5 from 0.41 → 0.70 by replacing basic keyword retrieval with hybrid BM25 + dense retrieval and metadata filtering.
  • Reduced unsupported LLM responses by 32% by implementing retrieved-context validation, source grounding, and response constraints across a 500-query evaluation dataset.
  • Improved retrieval quality to 0.70 Recall@5 and 0.62 MRR through iterative evaluation of chunk size, embedding strategy, metadata filtering, and top-K retrieval parameters.
  • Implemented LangGraph-based conversation state and memory management, maintaining context across 10+ conversational turns while restricting responses to retrieved legal knowledge.

Resume Ranker

LangChain, Python, NLP, Embeddings, LLMs
  • Built an NLP-based resume ranking pipeline extracting candidate skills, experience, education, and job requirements to generate relevance scores against job descriptions.
  • Created a labeled evaluation dataset of 100 resumes across 30 job descriptions and evaluated ranking quality using Precision@K, Recall@K and NDCG@K.
  • Improved NDCG@10 from 0.61 → 0.82 by combining semantic embedding similarity with structured skill and experience matching instead of relying solely on keyword overlap.
  • Improved Precision@10 from 0.68 → 0.84 through score normalization, skill-weight optimization, and semantic similarity-based candidate ranking.
  • Reduced candidate screening effort by 60%, ranking approximately 100 resumes in under 2 minutes compared with manual resume-by-resume screening.

Multi-Agentic AI Marketing System

CrewAI, LLMs, Streamlit
  • Developed a multi-agent AI marketing system using CrewAI to automate and optimize marketing strategies.
  • Integrated Ollama 3.1:8b (local) as the LLM to enhance decision-making and content generation.
  • Implemented intelligent agents for campaign management, audience targeting, and performance analysis.
  • Improved engagement and conversion rates through data-driven marketing strategies.
  • Deployed the system on Streamlit for seamless user interaction and accessibility.

Vectorization and Validation

Vector Databases, Azure Databricks, RAG
  • Developed an AI-powered Q&A system processing different types of documents for efficient query resolution.
  • Implemented ChromaDB & Mosaic AI for vector search, achieving a 70% improvement in retrieval accuracy.
  • Validated results using cosine similarity and Euclidean distance, achieving up to 92% query accuracy.
  • Deployed on Azure Databricks for scalability and real-time performance.

Medical Imaging AI System

BLIP + RAG + Vector Database
  • Developed a multimodal medical imaging system using BLIP (Bootstrapped Language-Image Pretraining) to detect abnormalities in chest X-ray images.
  • Designed and implemented a Retrieval-Augmented Generation (RAG) pipeline integrating FAISS/ChromaDB, enhancing contextual explanation quality and increasing diagnostic insight relevance by 25%.
  • Engineered an end-to-end workflow for image embedding generation, semantic retrieval, and structured report synthesis, reducing manual interpretation effort by 30–40%.
  • Deployed a scalable backend architecture for real-time inference and medical report generation, achieving sub-2-second response latency and improving processing efficiency by 40%.

Deep Learning Model Optimization & Finetuning (Image Generation)

PyTorch + Weights & Biases
  • Built and fine-tuned a deep learning model using PyTorch, optimizing hyperparameters through structured experimentation for domain-specific image generation and improving model accuracy by 12%.
  • Leveraged Weights & Biases (W&B) for experiment tracking, hyperparameter sweeps, and model comparison, reducing experimentation time by 35%.
  • Conducted systematic tuning of learning rate, batch size, optimizer configuration, and regularization techniques, reducing validation loss by 20%.
  • Implemented performance monitoring using accuracy, precision, recall, F1-score, and loss tracking, enabling reproducible training pipelines and improving model stability by 20–30%.

Soft Skills

  • Strong analytical and problem-solving skills with a structured approach to troubleshooting and root cause analysis.
  • Effective communication and stakeholder collaboration in cross-functional technical environments.
  • Ownership-driven mindset with a focus on reliability, accountability, and continuous improvement.
  • Ability to work under pressure in production environments while maintaining SLA commitments.
  • Adaptability and quick learning capability in fast-paced AI and cloud-native ecosystems.
Tanishq's avatar

Tanishq Tembhurne

Hyderabad, India

AI Engineer making
production systems fix themselves faster.

AI SRE Engineer · AI Agents & LLMs · Azure Observability

I build reliable cloud workflows and AI-powered systems — from automated incident investigation and observability to RAG applications, multi-agent tools, and machine learning projects.

ai-sre-agent · investigating
INC-2041
Animated illustration of an AI SRE agent investigating production alerts: it queries telemetry, matches a runbook, drafts a root-cause analysis and posts it to Jira for human review.

30%

lower MTTR with AI-assisted investigation

50%

less manual incident-management effort

30+

production alerts auto-investigated / month

3

Azure production environments covered

01 — Journey

From data to AI agents in production

Each step built on the last: data analysis taught me to measure, ML taught me to model, and SRE taught me what it takes to run AI reliably.

  1. 2021

    Started Computer Science · Woxsen University, Hyderabad

    B.Tech in Computer Science Engineering (2021 – 2025). Built my foundations in Python, ML and data.

  2. Feb – Jun 2024

    Data Science Intern · Conquest Technology Solutions

    Turned messy datasets into Power BI KPI dashboards. Learned to answer business questions with data.

  3. Projects

    From models to agents

    Moved from deep learning and computer vision to RAG, knowledge graphs and multi-agent systems — measuring every step with real evaluation sets.

  4. Apr 2025 – Now

    SRE Engineer · Centific

    Bringing LLMs into production operations: agentic incident investigation, telemetry correlation and Jira automation on Azure.

What I focus on

  • Site Reliability Engineering & incident response
  • Azure observability & automation
  • AI-assisted root-cause analysis
  • Agentic workflows & multi-agent systems
  • RAG & semantic search
  • NLP, deep learning & computer vision

02 — Experience

Where I’ve worked

Centific

SRE Engineer

April 2025 – Present

30%

MTTR reduction

50%

less manual effort

30+

alerts / month automated

Reliability, observability and automation across Azure production environments — helping responders investigate incidents faster with AI.

  • Engineered an AI-assisted incident investigation system across 3 Azure production environments, producing evidence-backed root-cause analyses.
  • Designed an agentic workflow that reads alerts, queries telemetry with KQL, retrieves runbooks and history, and writes structured RCAs.
  • Automated alert-to-investigation flows with Azure Monitor, Logic Apps and Jira — from kickoff to RCA publication and incident updates.
  • Designed a human-in-the-loop AI SRE workflow with progressive RCA updates aligned to a 30-minute incident SLA.
  • Built email-to-Jira automation with Logic Apps, Jira REST APIs, SharePoint thread mapping and a Node.js Azure Function.
  • Azure Monitor
  • App Insights
  • Log Analytics
  • KQL
  • Logic Apps
  • Azure Functions
  • Jira API
  • Grafana
  • LLMs

Conquest Technology Solutions

Data Science Intern

February 2024 – June 2024

15%

sales lift in low-performing areas

Exploratory analysis and KPI dashboards that supported business decisions.

  • Developed and analysed KPI dashboards in Power BI, including client-oriented reporting work.
  • Ran exploratory data analysis to surface trends, patterns and anomalies.
  • Worked with stakeholders to make reports more accurate and easier to use.
  • Power BI
  • Python
  • Pandas
  • EDA

03 — Projects

Things I’ve built

Each card shows the problem, how I solved it and what it achieved. Open “What I built” for details.

Work · Centific

AI SRE — Incident Investigation Agent

On-call engineers lose time stitching logs, metrics, traces and runbooks together before they can even start fixing an incident.

What I built
  • Agent reads the alert, writes KQL to query production telemetry, and pulls matching runbooks and past incidents.
  • Correlates logs, metrics, traces, dependencies and deployments into one evidence-backed RCA.
  • Publishes findings and progressive updates to Jira, with a human reviewing before closure.
  • Azure Monitor
  • KQL
  • Logic Apps
  • Jira
  • LLM reasoning
lower MTTR
30%
lower MTTR
less manual effort
50%
less manual effort
RCA update SLA
30 min
RCA update SLA

Real-time voice agent

Autonomous Meeting Agent

Meetings need someone to listen, answer questions, track decisions and send the minutes.

What I built
  • A virtual employee that joins Chromium meetings, listens, asks and answers questions, and presents content.
  • Fine-tuned small language models; tuned chunking, embeddings, metadata filters and top-K for retrieval.
  • Short- and long-term memory for decisions and action items, plus Redis caching for fast answers.
  • Python
  • Node.js
  • SLMs
  • RAG
  • OpenAI Realtime
  • ElevenLabs
  • Redis
Recall@5
0.52 → 0.76
Recall@5
smaller prompts
45%
smaller prompts
lower latency
40%
lower latency

Graph + vector RAG

Legal-Aid Conversational Chatbot

Legal answers must stay grounded in the source documents — a guess is worse than no answer.

What I built
  • Hybrid retrieval: BM25 + dense vectors in FAISS, with relationships from a Neo4j knowledge graph.
  • Context validation, source grounding and response constraints to cut unsupported answers.
  • LangGraph state for multi-turn memory across long conversations.
  • LangChain
  • LangGraph
  • FAISS
  • Neo4j
  • Groq
Recall@5
0.41 → 0.70
Recall@5
fewer unsupported answers
32%
fewer unsupported answers
turns of context
10+
turns of context

More projects

NLP · Ranking

Resume Ranker

Screening resumes one by one is slow and inconsistent.

What I built
  • Extracts skills, experience and education from resumes and job descriptions.
  • Blends embedding similarity with structured skill and experience matching.
  • Evaluated on 100 resumes × 30 job descriptions with Precision@K, Recall@K and NDCG@K.
  • Python
  • LangChain
  • Embeddings
  • LLMs
NDCG@10
0.61 → 0.82
NDCG@10
less screening effort
60%
less screening effort

Agentic AI

Multi-Agent Marketing System

Campaign planning, targeting and analysis usually live in separate tools and heads.

What I built
  • Specialised CrewAI agents for campaign management, audience targeting and performance analysis.
  • Runs on a local Ollama 3.1:8b model for decisions and content generation.
  • Streamlit interface for interacting with the agent crew.
  • CrewAI
  • Ollama
  • LLMs
  • Streamlit

Document Q&A

Vectorization & Validation

Answering questions across many document types at scale.

What I built
  • Vector search with ChromaDB and Mosaic AI.
  • Compared cosine similarity and Euclidean distance for retrieval.
  • Deployed the Q&A workflow on Azure Databricks.
  • ChromaDB
  • Mosaic AI
  • Azure Databricks
  • RAG
better retrieval accuracy
70%
better retrieval accuracy
query accuracy
92%
query accuracy

Multimodal · Experimental

Medical Imaging AI

Exploring how vision-language models can draft structured reports from chest X-rays.

What I built
  • BLIP for image understanding, combined with RAG over FAISS / ChromaDB.
  • Image embeddings, semantic search and structured report synthesis.
  • Backend for real-time inference and report generation.
  • BLIP
  • RAG
  • FAISS
  • ChromaDB

Research project — not a clinically validated diagnostic tool.

response latency
<2s
response latency
less manual interpretation
30–40%
less manual interpretation

Deep learning

Image Generation Fine-Tuning

Getting a stable, domain-specific image model without endless trial and error.

What I built
  • Fine-tuned a PyTorch model for domain-specific image generation.
  • Structured sweeps over learning rate, batch size, optimizer and regularization.
  • Weights & Biases for experiment tracking and model comparison.
  • PyTorch
  • Weights & Biases
higher accuracy
12%
higher accuracy
faster experiments
35%
faster experiments
lower val. loss
20%
lower val. loss

Figures are as reported on my resume and in each project's own evaluation.

04 — Skills

Toolkit

Agentic AI & Retrieval

  • RAG
  • LangChain
  • LangGraph
  • LlamaIndex
  • CrewAI
  • Tool calling
  • Hybrid search
  • Reranking
  • FAISS
  • ChromaDB
  • Neo4j

AI / Machine Learning

  • LLMs & SLMs
  • Fine-tuning
  • PyTorch
  • NLP
  • Computer Vision
  • Model evaluation

Cloud, SRE & Automation

  • Azure
  • AWS
  • GCP
  • Azure Monitor
  • App Insights
  • KQL
  • Logic Apps
  • Azure Functions
  • Databricks
  • Docker
  • CI/CD

Languages & Data

  • Python
  • JavaScript / Node.js
  • Pandas
  • NumPy
  • Power BI
  • Matplotlib

05 — Contact

Building something with AI agents or reliability? Let's talk.