Projects

Case studies

AI systems and software products — the problem, how I approached it, the architecture and the measured results.

01 / 2025 · AI/ML · Data · Full-Stack

Product Carbon Footprint Engine

Fine-tuned LLMs read a manufacturer's raw parts list, match every material against a 1M+ record emissions library and produce the product's carbon footprint report — automatically.

98.7%
Parts-list parsing accuracy
1M+
Materials searchable
40%
Lower retrieval latency
How it works
01

A manufacturer uploads a messy, multi-page parts list

System pipeline · for engineers
Parts-list uploadasync parsing pipelinelocal LLMsstructured materials
Materialshybrid Qdrant search (vectors + metadata filters)LCA library match
Fuzzy match + multi-LLM voteconfidence scorecarbon footprint calculation
Flask APINDJSON / SSE streamReact UI
PythonFlaskQdrantLoRA / QLoRALLMsSQLSSE

02 / 2026 · AI/ML · Full-Stack

AnswerGate — Regression Tests for AI Assistants

An open-source CI check that catches the moment an AI assistant's answers get worse after a code change, and fails the pull request with the exact questions that got worse.

36/42
Injected faults caught
0
False alarms, 1,070 questions
19/44
Questions flagged by one check

Watch · 1:35 · sound on

Pull request · “Smaller chunks for more precise retrieval”

examples/fastapi_qdrant/rag.py

−CHUNK_SIZE = 1000 # characters per chunk

+CHUNK_SIZE = 250 # characters per chunk

One line. Looks harmless in review.

How it works
01

A pull request changes one line in an AI assistant

System pipeline · for engineers
Pull requestGitHub ActionAnswerGateyour AI assistant (HTTP or Python)
Answers + retrieved passagesretrieval metrics + local LLM judgeevidence check
Scoresbaseline diff + 95% confidencepass / fail gate
VerdictPR comment + HTML report
PythonTyperPydanticOllamaQdrantFastAPIGitHub Actionspytest

03 / 2026 · Full-Stack

Exsy — Family Money Book

An installable money app for a family: separate books per person, EMIs with GST, loans, a chit fund with fair draws, and entries filled in from a pasted bank SMS.

Watch · 1:03 · sound on

System pipeline · for engineers
Phone or laptopinstallable PWA (vanilla JavaScript, Material 3, no build step)works offline via a service worker
Entriesderived balances, budgets, EMI schedules, chit statusHome, reports, "What this means" explanations
Google sign-inFirestore users/{uid} (security rules: owner only)live sync across devices
StatementsjsPDFShare sheetWhatsApp reminder with the PDF attached
JavaScriptPWAFirebase AuthFirestoreMaterial 3jsPDF

04 · AI/ML · Data

Adaptive Traffic Signaling — with Bangalore Traffic Police

LSTM-based traffic optimisation engine that learns from high-frequency, real-time sensor streams to adapt signal timing, built in collaboration with Bangalore Traffic Police.

30%
Less congestion (simulated)
How it works
01

Cameras detect and count every vehicle at each approach

System pipeline · for engineers
Traffic sensorsreal-time ingestion pipelineSQL store
Recent flow historyLSTM forecastersignal-timing optimiser
PythonLSTM Neural NetworksSQLData Pipelines