Pranav Kolle
AI Product Engineer building AI products that automate real business workflows — AI-powered SaaS, intelligent automation systems, and the integration layers that hold them together.
01 — How I think
Before writing code, I focus on understanding the business workflow. My goal is to reduce manual work through automation, simplify complex processes with AI, and build software that teams actually enjoy using. Every project starts with identifying the bottleneck — not the technology.
Currently Forward Deployed Engineer at SKYOVI, embedded with clients to turn a vague business requirement into a system that runs in production. Before that, four years building enterprise software at Tech Mahindra — which is where I learned that the hard part is almost never the algorithm.
02 — What I build
Four shapes of problem
Different domains, same underlying job: find the step a business does by hand, and design the system that removes it.
- 01
AI-powered SaaS
End-to-end products where the AI is the core of the value, not a feature bolted onto a CRUD app — including billing, analytics and the unglamorous operational surface around them.
- 02
Workflow automation
Finding the step a team dreads doing by hand and removing it. Usually document handling, data reconciliation, or reporting that someone rebuilds every Monday.
- 03
Enterprise integration
Custom integration layers between client infrastructure and modern systems. Embedded with the client, defining the requirement and shipping against it.
- 04
Document intelligence
OCR, extraction, validation and comparison over messy real-world documents — including multilingual input where off-the-shelf tooling gives up.
03 — Selected work
Engineering intelligence into functional products
Two of these have full case studies — the problem, the architecture, the decisions I'd defend, and the ones I'd revisit.
System Architecture Map
ZIPLINK
Telecom Telemetry Platform • 2023
A WiFi scan operations and telemetry platform for a major telecom client. Real-time KPI dashboards and a geocoded service map, plus an automated GenAI daily operations briefing that replaced a manual leadership report.
Legal Doc Verification
Document Intelligence • 2023
AI-assisted verification for legal and banking workflows. OCR across English and Marathi, structured comparison between document versions, AI validation of extracted fields, and generated discrepancy reports.
ERP & Business Automation
Modular Enterprise Platform • 2023
A modular business management system spanning CRM, lead management, ticketing, inventory, subscriptions, contracts and company administration — with workflow automation cutting across all of them.
04 — AI & system architecture
Predictable systems around unpredictable models
The same few pipeline shapes turn up across most of my work. These are the ones I keep reaching for.
Deterministic core
Build predictable infrastructure around non-deterministic models. State, sequencing and guarantees belong in ordinary code — the model handles language.
Grounding over model choice
The gap between a generic prompt and one carrying real context is wider than the gap between any two frontier models. Invest in the context layer.
Swappable providers
Pricing, latency and availability all move. Every LLM call goes through one internal interface so switching is a config change, not a refactor.
Humans stay in the loop
AI drafts, flags and proposes. On anything consequential a person still approves — which is also what makes the output trustworthy enough to use.
Hover or focus a stage to see why it’s built that way.
Hover or focus a stage to see why it’s built that way.
Hover or focus a stage to see why it’s built that way.
05 — Experience
Enterprise systems → AI products
Evolution
Enterprise foundations
2021Started in high-throughput enterprise systems. Learned what actually breaks at scale, and that most of it is not the interesting part of the codebase.
Integration & architecture
2022Moved toward the seams between systems — API design, data pipelines, and the runbooks that keep a handoff from becoming an outage.
Automation
2023Began replacing manual business processes outright rather than building interfaces for them. OCR, document workflows, telemetry reporting.
AI in production
2024Shipped LLMs into real client workflows — contextual AI pipelines, voice agents, GenAI operations briefings. Deterministic infrastructure around non-deterministic models.
AI product engineering
TodayBuilding AI-first SaaS end to end: finding the bottleneck in a business workflow, then designing the product and the system that removes it.
Technical Arsenal
AI & LLMs
- OpenAI API
- Claude API
- Prompt Engineering
- GenAI MCP
- AI Agents
- Vector Embeddings
- OCR
- Speech APIs
Backend & Infra
- Node.js
- Express
- Spring Boot
- Python
- PostgreSQL
- MySQL
- MongoDB
- Docker
- Kubernetes
Frontend & Interface
- React
- Next.js
- Vue.js
- TypeScript
- Tailwind CSS
- Three.js
- WebSockets
SKYOVI
Aug 2024 — PresentForward Deployed Engineer, Full Stack
Embedded directly with startup clients to define technical requirements and architect custom integration layers.
- Delivered 4 client projects with 100% customer satisfaction
- Boosted client operational and onboarding efficiency by 35% with production features in React, Next.js and Spring Boot
- Cut client data retrieval times by 20% through optimised RESTful data pipelines over MongoDB and MySQL
- Designed contextual AI workflows and custom API connectors that automate manual business processes
Tech Mahindra
Nov 2021 — Jul 2024Software Engineer, Enterprise Solutions
End-to-end design and implementation of high-throughput React and Node.js applications for enterprise clients.
- Patched and refactored critical legacy codebases, improving core system stability and UI responsiveness
- Authored technical blueprints, API architecture diagrams and system runbooks for engineering handoffs
- Aligned delivery timelines across product, QA and external operations teams
Education & certification
PG Diploma in Advanced Computing (PG-DAC)
2021C-DAC, Pune
B.E. Computer Science
2020VDF College of Engineering & Technology, Latur — GPA 9.2 / 10
Advanced Crash Course in GitHub Copilot & OpenAI Systems
2024Certification
06 — Building in public
Smaller things I'm poking at
Experiments that exist to answer one question. Some of them turn into products — the review engine below became Reewu.
Prompt Engineering
AI Review Generation Engine
The prompt-engineering core that became Reewu. Explores how much business context — category, services, location, customer intent — has to be injected before generated text stops reading as generic.
Prompt DesignContext GroundingEntity Resolution
AI Contact Matching
Contact normalisation and duplicate detection across multiple data sources, using vector embeddings to catch the near-matches that exact and fuzzy string rules miss.
Vector EmbeddingsDeduplicationExperiment
AI Reply Generator
Reads conversation context and drafts a response in a chosen register — Friendly, Apology, Debate, Professional — while holding onto the user's own voice rather than flattening it into assistant-speak.
Tone TransferContext Windows
07 — Contact
Let’s build something.
Open to AI product work, consulting on LLM systems, and full-stack engineering roles. If you have a workflow that eats your team’s week, I’d like to hear about it.