Leapfrogs '26 · BTH × Lund × HKR · Karlskrona, SE

A used-car guide
for the carbon market.

I'm Abhinav Annareddy. I'm building CarbonLens — an AI-powered fair-value calculator for the Voluntary Carbon Market. It tells corporate buyers if the price a broker just quoted is reasonable. Nobody else does this.

Now Founder, CarbonLens
Stage MVP shipping August 2026
Showcase TechBBQ Copenhagen, Sept 2026
Status Open to conversations
01 Live data EU ETS spot CarbonLens API
Live · Updated May 9, 2026 — 14:22 CET
0.42
24h change +2.3%
Fair values today 156
Avg. confidence 87.4%
Sentiment Bullish
02 Why this exists
The voluntary carbon market is worth two billion dollars a year, growing to fifty by 2030. Companies are pouring money into offsets and have no honest way to know if they're paying a fair price. CarbonLens fixes that.
A.A. · Founder's note, May 2026
03 Selected work

An index of work.

2026
CarbonLens AI-powered fair-value engine for voluntary carbon credits. Founder, full-stack.
PyTorchXGBoostNext.jsFinBERT
Shipping
2025
LSTM equity forecasting Time-series prediction with walk-forward validation. The architectural ancestor of CarbonLens.
PyTorchNumPy
Archive
2024
Viak Group data pipelines Batch error rate from 12% to under 1%. Reporting time from 4 hours to 15 minutes.
AzurePythonPower BI
Industry
2024
Decision-support regression XGBoost + SHAP for operations. Surfaced three levers that weren't being measured.
XGBoostSHAP
Industry
2023
Bedrock RAG assistant Retrieval pipeline mixing OpenAI embeddings and LLaMA 2. Internal knowledge tool, ran on EKS.
LangChainAWS Bedrock
Industry
04 About

Computer science by training. ML by trade. Climate fintech by conviction.

I'm a Leapfrogs 2026 scholar at BTH in Karlskrona, Sweden — a cross-institutional programme with Lund University and HKR. My thesis is on applying machine learning to fair-value pricing in the Voluntary Carbon Market, a space where opacity is the norm and data is scarce.

Before this, I spent two years building data pipelines and ML models in industry — reducing batch error rates from 12% to under 1%, building RAG systems on AWS Bedrock, and shipping SHAP-driven decision tools that surfaced operational levers nobody was measuring.

Education
BTH (MSc)Lund UniversityHKRGITAM (BTech)
Models
PyTorchXGBoostFinBERTLSTMSHAP
Backend
PythonFastAPIPostgreSQLRedis
Frontend
Next.jsTypeScriptAstroReact
Cloud
AWSAzureDockerVercel
05 The ask

Let's talk carbon.

abhinav@carbonlens.app