Skip to content
View AswaniSahoo's full-sized avatar
:octocat:
On My Way To Find
:octocat:
On My Way To Find
3 followers · 27 following

Achievements

Achievement: Pull Sharkx2Achievement: Pair ExtraordinaireAchievement: YOLOAchievement: Quickdraw

Achievements

Achievement: Pull Sharkx2Achievement: Pair ExtraordinaireAchievement: YOLOAchievement: Quickdraw

Highlights

  • Pro

Block or report AswaniSahoo

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
AswaniSahoo/README.md

ASWANI SAHOO

Typing SVG

Merged PRs PyTorch Twitter LinkedIn

About

I build ML systems for science and production: LLM agents with measured evaluation, weather forecasting models, genomics NLP.

Not from a CS background. Taught myself ML, shipped code upstream, earned PyTorch org recognition. Most of what I know came from having patches reviewed by maintainers who did not have to be kind about it.

B.Tech @ NIT Rourkela '27 | CGPA 8.02 | Odisha, India

Open Source: 33 Merged PRs across 5 organizations

  • Built the adaptive-mesh regional forecasting stack end to end: DynamicGraphBuilder, RegionalForecaster, BoundaryNudgingLayer, variable-resolution H3 mesh, stretched-grid dataset and region-weighted loss
  • Root-caused a silent failure where the decoder discarded the encoder's per-observation features, capping the model at the no-change baseline. Skip-connection fix, zero new parameters, held-out skill 4% → 20% over persistence
  • Earlier: ThermalizerLayer diffusion denoising, NNJA-AI V1 dataset loader
  • Listed as contributor in graph_weather v1.0.132 (Zenodo DOI)

  • Docathon 2026: Honorable Mention (first-timer category)
  • Migrated 5 core doc files (nn.functional, autograd, extending.func, mkldnn, gradcheck) from RST to MyST
  • Fixed broken {include} paths rendering 3 empty backend pages
  • Documented the CMake build + ctest workflow for C++ tests
  • Added an Inspector API usage example with print_data_tabular()

  • Fitness-window bug fix in the chaos-scenario scoring path
  • Dependency sync across the krkn-ai toolchain
  • Fixed missing weight field in network-chaos-ng scenario frontmatters

  • Added lower-triangle annotation for Fst heatmaps in the malaria vector genomics library
  • Fixed a CNV data-check indentation bug causing silent failures

Currently Active

  • graph_weather #3 / #238 — adaptive meshing for regional NWP. Currently running a controlled ablation on whether the mesh message-passing earns its parameters against a graph-free baseline; publishing the numbers either way.
  • krkn-chaos/krkn-ai #389 — Elasticsearch config for composite chaos scenarios
  • malariagen-data-python #1310 — haplotype network mixin

Projects

Project What it does Highlights
Climate-Risk Analyst Agent — live LangGraph agent returning typed, cited climate-risk reports — or refusing when the evidence is thin 105-question SHA-frozen held-out set: recall@3 87%, citation validity 94%, zero false answers. 238 tests, ~$0.001/report, deployed on Cloud Run
Incident Evidence Compiler Root-cause investigation service where the model proposes and deterministic code decides RCAEval RE2-OB: top-1 0.932, MRR 0.959. Content-addressed evidence ledger, Postgres SKIP LOCKED worker queue, 304 hermetic tests, mypy strict
Weather Transformer Physics-aware Transformer for 6-hour weather forecasting on ERA5 Written from scratch in PyTorch — no nn.MultiheadAttention. 74 tests, beats persistence by 27% RMSE on a 1,316-sample held-out set
Complaint Intelligence — live RAG and retrieval benchmark over CFPB consumer complaints 12M+ records processed, 200K partition deployed. MiniLM vs BGE, vector vs BM25 vs hybrid vs rerank — every component chosen on measured results
LLaMA Task Agent LoRA fine-tuned LLaMA-3.1-8B for agentic tool execution Generates valid function calls with type-safe args from natural language
Bio Publication Analyzer SciBERT + TF-IDF classifier for genomics articles 81 tests, Europe PMC pipeline end to end

Tech Stack

GitHub Stats

Pinned Loading

  1. climate-risk-agent climate-risk-agent Public

    Open-source agentic climate-risk analyst: grounded, cited, structured risk reports from live weather + IPCC docs.

    Python 1

  2. Incident-evidence-compiler Incident-evidence-compiler Public

    Python 1

  3. weather-transformer-scratch weather-transformer-scratch Public

    Physics-aware Vision Transformer for weather forecasting built from scratch in PyTorch | AI for Science

    Jupyter Notebook 1

  4. llama-task-agent llama-task-agent Public

    Fine-tuned LLaMA-3.1-8B task agent with LoRA for reliable tool execution

    Jupyter Notebook 1

  5. complaint-intelligence-system complaint-intelligence-system Public

    RAG pipeline for analyzing 12M+ customer complaints using FAISS, Sentence-Transformers & Gemini AI

    Jupyter Notebook 1

  6. fairness-credit-risk fairness-credit-risk Public

    Fairness-aware credit risk modeling with bias detection, AutoML & AI Fairness 360

    Python