Ganesh Kambli
AI Engineer at Flid AI
I build software end to end, from the first sketch to a product running in production.
At Flid AI I work on LeapView, open-source BI in Go and DuckDB, mostly the parts that have to be right: who can see which data, how fast queries come back, and whether a release installs and upgrades cleanly. I also take on freelance projects.
390 merged pull requests in LeapView since Aug 2026Work at Flid AI
LeapView
Define metrics in code. Explore them everywhere.
LeapView is open-source BI. Teams define a metric once in Git and use it in dashboards, APIs and AI tools, with the same access rules applied everywhere. The server is Go, queries run on DuckDB, and it is licensed Apache-2.0.
metric: merged_pull_requestssource: github.com/flidai/leapviewauthor: Ganesh-403grain: week 390 merged pull requests
20 Aug to 24 Oct 2026
Weekly numbers
| Week of | Merged |
|---|---|
| 17 Aug | 3 |
| 24 Aug | 34 |
| 31 Aug | 44 |
| 7 Sep | 43 |
| 14 Sep | 25 |
| 21 Sep | 33 |
| 28 Sep | 55 |
| 5 Oct | 131 |
| 12 Oct | 15 |
| 19 Oct | 7 |
Where my changes go
-
Releases and supply chain Reproducible Nix builds, signed provenance and SPDX SBOMs attached to every release candidate, and qualified Linux desktop installers.
-
Upgrades and recovery Backup and restore, a durable recovery ledger, and schema upgrades tested against real release candidates, so self-hosted installs can move forward safely.
-
Governed semantic layer Access grants compiled into typed row and column predicates, plus ODCS 3.1 export, OpenLineage and dbt interoperability.
-
Query performance Arrow streaming for governed queries and a result cache that keeps every access check in place.
-
Security and identity SCIM deactivation that revokes agent (MCP) sessions, credential lifecycle recovery, and a hardened Chainguard runtime image.
-
Reliability Service-level indicators and burn-rate alerts for dashboard refresh, W3C trace context, and the runbooks that go with them.
LeapView on GitHub leapview.dev All my merged pull requests
I also work on Alloy Catalog, Flid's engineering master-data catalog with a built-in AI agent.
Projects
-
Ask questions about any codebase, fully offline.
How it works: RepoSage
Splits code into functions and classes, maps which function calls which into a graph, and answers questions with a local model through LangGraph. Every answer cites the exact file and line. Retrieval combines ChromaDB vectors with BM25 keyword search.
Python, LangGraph, Ollama, ChromaDB, FastAPI, React
-
Finds paraphrased and cross-language copying in student work.
How it works: Semantic Plagiarism Detector
Embeds assignments with a multilingual sentence model, searches them with FAISS, and shows which sections match and how closely, down to a paragraph-level heatmap. I ran it as an open-source project for ECSoC’26.
Python, Sentence Transformers, FAISS, Streamlit
-
A static analyzer for application configuration.
How it works: envtether
Finds every environment variable a Python project reads, flags drift between .env files, Docker Compose and CI, and catches hardcoded secrets using patterns plus entropy checks. Reports in SARIF for GitHub code scanning.
Python, AST analysis, SARIF
-
Decoy servers that record, score and profile attackers.
How it works: HoneyCloud
Fake SSH, FTP, HTTP, Telnet, SMTP and RDP services capture attack traffic, score it with machine learning, map it to MITRE ATT&CK and stream it to a live dashboard. Rewritten in Rust; the research behind it became two papers.
Rust, Axum, Tokio, PostgreSQL, WebSockets
-
A virtual file system on a real disk image.
How it works: SVF
Formats a binary image into a superblock, block bitmaps and 4 KB data blocks, then serves files through inodes guarded by reader-writer locks, with POSIX permissions and Argon2id-hashed user accounts.
C++17, CMake, Argon2id
-
Huffman frequency counting on CPU, OpenMP and CUDA.
How it works: GPU Huffman
Benchmarks the same character-frequency count across a single CPU thread, OpenMP, and three CUDA kernels: global atomics, shared-memory histograms and warp-level aggregation, with throughput reports per run.
C++17, CUDA, OpenMP
Five earlier projects
- Flight Delay AI XGBoost delay predictions explained with SHAP, with weather cached in Redis and live updates over WebSockets.
- AWS URL shortener Serverless on Lambda, API Gateway, DynamoDB, S3 and CloudFront. Live demo for AWS URL shortener
- RailExp A legacy Flask railway system rebuilt as FastAPI and React, shipped with Docker.
- Aura Wellness An Android app in Kotlin that steps in before a doomscroll, with Gemini-written check-ins.
- Image colorization Colorizes black-and-white photos with a pretrained CNN, guided-filter upsampling and tiled high-resolution inference.
Freelance
I take on freelance projects alongside my role at Flid AI and see them all the way through: product decisions, design, code, deployment, and the fixes after launch.
-
A new product, from idea to launch
You bring the problem. I handle the rest: product decisions, interface, backend, data and hosting, and hand over something you can run.
Related work: RepoSageSemantic Plagiarism Detector
-
An existing system, taken further
New features, faster queries, AI connected to your own data, or fixes for the parts that keep breaking, without starting over.
-
Ready for real users
Releases you can repeat, upgrades that don’t lose data, access control and monitoring, so it keeps working after launch.
I work from Mumbai (IST, UTC+5:30). Tell me what you're building and when you need it, and I'll tell you honestly whether I'm the right fit.
Work with meBackground
Experience
- Aug 2026 to now AI Engineer Flid AI, remote from Mumbai Full-time LeapView: governed data access, query performance, releases and upgrades.
-
Feb 2025 to Mar 2025 Java Developer, Backend and Microservices ShadowFox, Pune Internship
- Built a microservices e-commerce backend with REST APIs and service boundaries that deploy independently.
- Delivered five Java applications, including bank and student management systems, test-first with JUnit.
-
Jan 2025 to Feb 2025 Python Developer, Automation and Data Systems AI Adventures, Pune Internship
- Built an absence detection system on the Google Sheets API that made attendance tracking over 30% more efficient.
- Automated data-change alerts with GitHub Actions, cutting manual follow-up by 90%.
- 2022 to 2026 B.E. Computer Engineering Genba Sopanrao Moze College of Engineering, Savitribai Phule Pune University Education CGPA 7.95
Open source
Outside LeapView, I send fixes upstream to projects I use and help run open-source programs for students.
- riscv/riscv-unified-db Populate long_name for LRSC, MISA, and trap parameters Sep 2026
- Apicurio/apicurio-registry Replace custom HSTSFilter with Quarkus-managed security headers Aug 2026
- riscv/riscv-unified-db Add missing Smctr CSR models and CTR depth parameter Aug 2026
- Project admin, Elite Coders Summer of Code Jul – Sep 2026 Ran the Semantic Plagiarism Detector as a program project: reviewed contributor pull requests, triaged issues, wrote contributor docs and issue templates, and set up branch protection.
- Contributor, GirlScript Summer of Code May – Aug 2026
Research
- A Survey on AI-Driven Honeypot Systems and Real-Time Intrusion Detection: Techniques, Challenges, and Future Directions
- HoneyCloud: A Smart Scalable Honeypot Platform with ML-Based Threat Classification and Real-Time Attacker Profiling
Awards and certifications 2 awards, 5 certifications
- Winner, Site Craft web development challenge VAMINT Club, 2025
- Runner-up, Prep-A-Thon coding competition Technobash, 2025
- Python for Machine Learning IIT Bombay (EduPyramids, SINE), 2026
- Google AI Essentials Google, 2025
- Gemini Certified University Student Google
- Introduction to Cybersecurity
- Learn Generative AI
I speak English, Hindi, Marathi and Konkani.
Contact
Tell me what you're building and where you're stuck. Email works best: