PMPranav Mulakala
~/pranav — zsh
pranav@bay-area ~ % whoami

Pranav Mulakala

Graduate Cloud Engineer · Sage Intacct, Data Platform & Reporting Co-founder & full-stack engineer · Whistle AI B.S. Computer Science · UC Santa Cruz · Dec 2025 pranav@bay-area ~ % ls ./building ai-agents/ automation/ data-platforms/ reliability/ pranav@bay-area ~ % open links github ↗linkedin ↗resume ↗ pranav@bay-area ~ %
How I work

I do my best work when I find a gap in how a team operates and build the automation that closes it.

At Sage that's meant AI agents in Slack, one registry of troubleshooting tools, and a provisioning runbook turned into a Slack flow. On my own it's a trading research platform, a robot-arm pipeline and an open-source fix awaiting review, each backed by its own tests or evals.

Results

Before and after, to scale

Each bar starts at the old number and shrinks to the new one as you scroll. Every bar has its own linear scale; hatched ends show where a reading was a range.

Ranking query latencySage · 2.8 TB dataset
45–55 s→1.3–2 s
OAS install completion timeSage · OAS Install Co-Pilot
4–5 h→30 min
Manual email-sorting timeMOURI Tech · GPT-4 + RAG classifier
100%→20%

Latency 0–60 s · install time 0–5 h · sorting time 0–100% of the manual baseline

At Sage · since Oct 2025Internal tools, so the code and screens are private. These are drawn in code with generic labels.
Incident thread
you
Incident report
A1
Specialist agent read-only
A2
Specialist agent read-only↳ picks up from A1
A3
Specialist agent read-only↳ picks up from A2
3 agents chained · no re-prompting10+ Snowflake Cortex agents
SageSnowflake award · hackathon

DPR Multiplayer AI

10+ Snowflake Cortex agents that work together in a Slack thread. One incident report chains three specialist agents with no re-prompting. I built nearly all of it and led our hackathon entry as its only engineer.

Snowflake CortexSlack
internal · private
One registry, three interfaces
Tool registry
22 tools
SlackCLIMCP
1,250+ automated testsSlack · CLI · MCP
Sage22 tools

DPR Command Center

One shared registry of 22 troubleshooting tools for infrastructure and the data platform, served the same way over Slack, a CLI, and MCP. I've presented my tools to senior engineers, PMs, and directors.

MCPSlackCLI
internal · private
Runbook → Slack flow
OAS install6 buttons replace 11 manual steps
4–5 h → 30 minresumable · keyless OIDC · audit-logged
Sagesolo-built

OAS Install Co-Pilot

Turned an 11-step AWS/Azure provisioning runbook into a 6-button Slack flow. Steps are resumable, AWS access uses keyless OIDC, and actions are audit-logged. The step grouping in the diagram is illustrative.

PythonAWSAzureGitHub ActionsAnsibleTerraform
internal · private
Ranking query, played at 10× speed
Before · slow ranking queries45–55 s
After · rollups + clustered tables1.3–2 s
2.8 TB datasetread-only SQL · access controls · freshness checks
Sage

Faster ranking queries

Replaced slow ranking queries over a 2.8 TB dataset with pre-aggregated rollups and clustered tables, then locked the path down with read-only SQL, access controls, and freshness checks.

SnowflakeSQL
internal · private
Health-check service · simplified, illustrative
Health eventscloud storage Kafkaproducers · consumers Workersretry on their own Redis cacherecent results Slack · Grafanaalerts · dashboards
Retriesfailed jobs, per worker
Cachesrecent results for Slack
Traceswhere slow commands wait
Sage

Health-check service

Background workers process cloud-storage health events from Kafka and retry failed jobs on their own. Redis caches recent results so repeated Slack requests come back fast. Grafana dashboards and Slack alerts flag services that keep failing, and OpenTelemetry traces show where a slow command spends its time.

KafkaRedisGrafanaOpenTelemetry
internal · private
Memory, illustrative
~29 daysfor the leak to build up
root cause: kernel memory leakowns 7 cloud-storage integrations
Sage

Incident work

Diagnosed a recurring outage caused by a kernel memory leak that built up over about 29 days. Separately, I own 7 cloud-storage integrations and led recovery from two fleet-wide incidents across them, including a credential failure affecting 15 production servers.

Linux
internal · private
Startup, open source and side projectsClips and screenshots come from the real apps. The Whistle diagram is drawn in code, and the terminal replays a real test run.
Annotation platform · abstract court, no real footage
Event17 fields
hotkey
HTTP range streamingffmpeg clips
Whistle AIco-founder

Basketball annotation platform

Turns game footage into labeled training data. ffmpeg cuts games into clips, video streams to the browser with HTTP range requests, and labelers tag each event with hotkeys against a fixed 17-field schema. My co-founder and I labeled the data and trained a YOLOv8 model on it together.

FastAPISQLAlchemyReact/ViteffmpegYOLOv8
private repo
▶ hover to play
Side projectpublic

Tape

A prototype that runs locally. It turns a stock's price tape into rated "plays", the way a sports app turns a game into plays, so everyone in a room would see the same move at the same moment. Shown: the live Bitcoin room on Coinbase's public feed.

TypeScriptWebSockets
Repository ↗
~/deterministic-swing-trading-research — zsh

              
            

            

              
Side projectpublic

              
Deterministic trading research

              
Paper-trading research built to be reproducible. A 13-stream market-data pipeline restarts from checkpoints, checks integrity with SHA-256, and writes to a hash-chained SQLite event journal. An Alpaca paper-trading adapter handles orders idempotently, reconciles state, and has kill switches.

              
PythonSQLiteLinuxAlpaca API

              
Repository ↗

            

          


          

            
▶ simulation
              
            

            

              
Side projectsimulation clip

              
WaterFilterRobo

              
Can a robot arm refurbish a used water purifier? I built a 5-stage Python pipeline on physical SO-101 robot arms for membrane handling, cleaning, and assembly, with a person verifying final cleanliness. Pilot partner: a leading water-purifier manufacturer. The clip is from the earlier MuJoCo simulation.

              
PythonMuJoCoLeRobot

              

            

          


          

            

                superglue agent proposing one of Pranav's run contracts
                A run contract flagging items that repeat an earlier id
                Live checks with two failures and fix hints
                Live checks with every subsystem passing
                Topology view with two injected failures
              

              
            

            

              
Built on open source5 screenshots

              
superglue diagnostics

              
Features I built on a local clone of superglue, an open-source project: REST/SSE diagnostics APIs in Node.js/TypeScript and a Next.js dashboard I demoed to superglue's co-founder. Run contracts detect data loss and duplicates. My changes are drafted as upstream contributions, not submitted yet. Click the stack to flip through the screenshots.

              
TypeScriptNode.jsNext.jsPostgreSQLDocker

              
superglue (upstream repo) ↗

            

          


          

            
▶ hover to play
              
            

            

              
Side projectpublic

              
BrokerDocs

              
A clickable React/TypeScript mockup of a freight-broker document tool (rate confirmations and bills of lading), built for customer-discovery demos.

              
ReactTypeScriptTailwind CSS

              
Repository ↗

            

          


          

            
testcontainers-python pull request 1144
              
            

            

              
Open sourcePR open · awaiting review

              
testcontainers-python: proposed fix

              
A proposed fix so HttpWaitStrategy applies response-body checks to accepted HTTP error statuses like 404 and 503. Built with AI assistance; adds 12 unit cases and 3 Docker integration cases.

              
PythonDockerpytest

              
PR #1144 ↗Issue #1143 ↗

            

          

        

      

    

  


  
  

    

      
Experience
The commit history
Newest first. Whistle AI is a separate venture I co-founded, so it sits on its own branch.

      
    
        
        
  1. 
          Oct 2025 – now
          
    Graduate Cloud Engineer
    
          Sage · Sage Intacct, Data Platform & Reporting · San Jose, CA
          
    AI agents, automation and data-platform tools for Sage Intacct's Data Platform & Reporting team. The Sage tiles above are the highlights.
    
        
    2. 
        
  2. 
          Dec 2025 – now
          
    Co-founder & full-stack engineer
    
          Whistle AI · Remote
          
    Built the annotation platform and data pipeline, and ran customer discovery.
    
        
    3. 
        
  3. 
          Sep 2022 – Dec 2025
          
    B.S. Computer Science
    
          University of California, Santa Cruz · GPA 3.81 · Dean's Honors 6 times
        
    4. 
        
  4. 
          Jun – Sep 2025
          
    Product Analyst (Intern)
    
          Okta · Anaplan Engineering · San Francisco, CA
          
    Automated Boomi-to-Anaplan bookings reporting for FP&A with a Python/AWS pipeline using Gemini-directed transformations over 1 TB of data, saving 3 days of manual reporting. Validated transfers against scenario-based tests before production, and added error summaries and an FP&A chatbot.
    
        
    5. 
        
  5. 
          Jun – Aug 2024
          
    Software / Data Engineer Intern
    
          MOURI Tech · Irving, TX
          
    Deployed a Python, GPT-4, and RAG classifier that sorts accounts-payable email: 85% precision and 80% less manual sorting time in timed comparisons.
    
        
    6. 
      

    

  


  
  

    

      
Toolbox
Stack and certifications

      

        

          
Languages
PythonSQLTypeScriptJavaScriptJavaGoC/C++C#Bash

          
AI & LLM
Snowflake CortexMCPOpenAI APIGeminiAWS BedrockRAGLangChainLangGraphPyTorchYOLOv8

          
Backend
FastAPINode.jsReactSQLAlchemyRESTKafkaRedisPostgreSQL

          
Cloud & infra
AWSAzureGCPSnowflakeDockerKubernetesTerraformAnsibleGitHub ActionsLinux

          
Observability
GrafanaPrometheusOpenTelemetryDatadogSplunk

          
Testing
pytestJestPlaywright

        

        

          

            
AWS Certified AI Practitioner2026

            
Microsoft Azure Fundamentals (AZ-900)2025

            
AWS Certified Cloud Practitioner2023

          

          
UC Santa Cruz · B.S. Computer ScienceDec 2025 · GPA 3.81 · Dean's Honors 6 timesData Structures · Object-Oriented Programming · Machine Learning · Artificial Intelligence · Cloud Engineering

        

      

    

  


  
  

    

      
Off the clock
What I'm into

      

        
TrainingLifting, four years in
I've lost about 100 pounds over the last three and a half years. Sticking with the gym built the discipline I bring to my work and to what I build on the side.

        
SportsBasketball
Through Whistle AI, the startup I co-founded.

        
MarketsTrading systems
I build my own market tools: the trading research platform and Tape.

      

    

  


  
  

    

      Contact
      
Let's build
something.

      
Email is the fastest way to reach me.

      

        
        
        Resume
      

      

        GitHubgithub.com/PranavMulakala
        LinkedInlinkedin.com/in/pranav-mulakala
        Emailpranavm1104@gmail.com
        Resumeone page · PDF
      

      
Bay Area · --:--

    

  







Resume