AswaniSahoo AswaniSahoo

ASWANI SAHOO

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-windows 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
Fairness-Aware Credit Risk	Four bias-mitigation interventions measured against a tuned baseline under identical conditions, on two credit datasets	No intervention helped — and the reasons differ per dataset. Bootstrap intervals on every metric, 212 tests, SHAP adverse-action reason codes, model card. Includes a published self-audit of three defects in my own earlier version, including a proxy leak where dropping `gender` did not remove sex from the model. Every published number regenerated from an artifact, enforced by a test
Complaint Intelligence — live	RAG and retrieval benchmark over CFPB consumer complaints	200K CFPB complaints processed end to end. MiniLM vs BGE, vector vs BM25 vs hybrid vs rerank, KMeans vs BERTopic — every component chosen on measured results, not defaults
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

ASWANI SAHOO

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-windows 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
Fairness-Aware Credit Risk	Four bias-mitigation interventions measured against a tuned baseline under identical conditions, on two credit datasets	No intervention helped — and the reasons differ per dataset. Bootstrap intervals on every metric, 212 tests, SHAP adverse-action reason codes, model card. Includes a published self-audit of three defects in my own earlier version, including a proxy leak where dropping `gender` did not remove sex from the model. Every published number regenerated from an artifact, enforced by a test
Complaint Intelligence — live	RAG and retrieval benchmark over CFPB consumer complaints	200K CFPB complaints processed end to end. MiniLM vs BGE, vector vs BM25 vs hybrid vs rerank, KMeans vs BERTopic — every component chosen on measured results, not defaults
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

ASWANI SAHOO

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-windows 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
Fairness-Aware Credit Risk	Four bias-mitigation interventions measured against a tuned baseline under identical conditions, on two credit datasets	No intervention helped — and the reasons differ per dataset. Bootstrap intervals on every metric, 212 tests, SHAP adverse-action reason codes, model card. Includes a published self-audit of three defects in my own earlier version, including a proxy leak where dropping `gender` did not remove sex from the model. Every published number regenerated from an artifact, enforced by a test
Complaint Intelligence — live	RAG and retrieval benchmark over CFPB consumer complaints	200K CFPB complaints processed end to end. MiniLM vs BGE, vector vs BM25 vs hybrid vs rerank, KMeans vs BERTopic — every component chosen on measured results, not defaults
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

ASWANI SAHOO

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-windows 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
Fairness-Aware Credit Risk	Four bias-mitigation interventions measured against a tuned baseline under identical conditions, on two credit datasets	No intervention helped — and the reasons differ per dataset. Bootstrap intervals on every metric, 212 tests, SHAP adverse-action reason codes, model card. Includes a published self-audit of three defects in my own earlier version, including a proxy leak where dropping `gender` did not remove sex from the model. Every published number regenerated from an artifact, enforced by a test
Complaint Intelligence — live	RAG and retrieval benchmark over CFPB consumer complaints	200K CFPB complaints processed end to end. MiniLM vs BGE, vector vs BM25 vs hybrid vs rerank, KMeans vs BERTopic — every component chosen on measured results, not defaults
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

ASWANI SAHOO

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-windows 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
Fairness-Aware Credit Risk	Four bias-mitigation interventions measured against a tuned baseline under identical conditions, on two credit datasets	No intervention helped — and the reasons differ per dataset. Bootstrap intervals on every metric, 212 tests, SHAP adverse-action reason codes, model card. Includes a published self-audit of three defects in my own earlier version, including a proxy leak where dropping `gender` did not remove sex from the model. Every published number regenerated from an artifact, enforced by a test
Complaint Intelligence — live	RAG and retrieval benchmark over CFPB consumer complaints	200K CFPB complaints processed end to end. MiniLM vs BGE, vector vs BM25 vs hybrid vs rerank, KMeans vs BERTopic — every component chosen on measured results, not defaults
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

ASWANI SAHOO

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-windows 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
Fairness-Aware Credit Risk	Four bias-mitigation interventions measured against a tuned baseline under identical conditions, on two credit datasets	No intervention helped — and the reasons differ per dataset. Bootstrap intervals on every metric, 212 tests, SHAP adverse-action reason codes, model card. Includes a published self-audit of three defects in my own earlier version, including a proxy leak where dropping `gender` did not remove sex from the model. Every published number regenerated from an artifact, enforced by a test
Complaint Intelligence — live	RAG and retrieval benchmark over CFPB consumer complaints	200K CFPB complaints processed end to end. MiniLM vs BGE, vector vs BM25 vs hybrid vs rerank, KMeans vs BERTopic — every component chosen on measured results, not defaults
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

ASWANI SAHOO

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-windows 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
Fairness-Aware Credit Risk	Four bias-mitigation interventions measured against a tuned baseline under identical conditions, on two credit datasets	No intervention helped — and the reasons differ per dataset. Bootstrap intervals on every metric, 212 tests, SHAP adverse-action reason codes, model card. Includes a published self-audit of three defects in my own earlier version, including a proxy leak where dropping `gender` did not remove sex from the model. Every published number regenerated from an artifact, enforced by a test
Complaint Intelligence — live	RAG and retrieval benchmark over CFPB consumer complaints	200K CFPB complaints processed end to end. MiniLM vs BGE, vector vs BM25 vs hybrid vs rerank, KMeans vs BERTopic — every component chosen on measured results, not defaults
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

ASWANI SAHOO

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-windows 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
Fairness-Aware Credit Risk	Four bias-mitigation interventions measured against a tuned baseline under identical conditions, on two credit datasets	No intervention helped — and the reasons differ per dataset. Bootstrap intervals on every metric, 212 tests, SHAP adverse-action reason codes, model card. Includes a published self-audit of three defects in my own earlier version, including a proxy leak where dropping `gender` did not remove sex from the model. Every published number regenerated from an artifact, enforced by a test
Complaint Intelligence — live	RAG and retrieval benchmark over CFPB consumer complaints	200K CFPB complaints processed end to end. MiniLM vs BGE, vector vs BM25 vs hybrid vs rerank, KMeans vs BERTopic — every component chosen on measured results, not defaults
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

AswaniSahoo AswaniSahoo

Highlights

ASWANI SAHOO

About

Open Source: 33 Merged PRs across 5 organizations

Currently Active

Projects

Tech Stack

GitHub Stats

Pinned Loading

Uh oh!

AswaniSahoo AswaniSahoo

Highlights

ASWANI SAHOO

About

Open Source: 33 Merged PRs across 5 organizations

Currently Active

Projects

Tech Stack

GitHub Stats

Pinned Loading

Uh oh!

AswaniSahoo AswaniSahoo

Highlights

ASWANI SAHOO

About

Open Source: 33 Merged PRs across 5 organizations

Currently Active

Projects

Tech Stack

GitHub Stats

Pinned Loading

Uh oh!

AswaniSahoo AswaniSahoo

Highlights

ASWANI SAHOO

About

Open Source: 33 Merged PRs across 5 organizations

Currently Active

Projects

Tech Stack

GitHub Stats

Pinned Loading

Uh oh!

AswaniSahoo AswaniSahoo

Highlights

ASWANI SAHOO

About

Open Source: 33 Merged PRs across 5 organizations

Currently Active

Projects

Tech Stack

GitHub Stats

Pinned Loading

Uh oh!

AswaniSahoo AswaniSahoo

Highlights

ASWANI SAHOO

About

Open Source: 33 Merged PRs across 5 organizations

Currently Active

Projects

Tech Stack

GitHub Stats

Pinned Loading

Uh oh!

AswaniSahoo AswaniSahoo

Highlights

ASWANI SAHOO

About

Open Source: 33 Merged PRs across 5 organizations

Currently Active

Projects

Tech Stack

GitHub Stats

Pinned Loading

Uh oh!

AswaniSahoo AswaniSahoo