contribution graph · regenerated every 12h by a scheduled Action
▼ show more ▼
$ whoami --verbose
ankit373 · Ankit Jha
├─ role Founding Engineer @ Tradomate
├─ before Tiger Analytics · Quantiphi ×2 · Wipro
├─ does platform & cloud architecture · Go · cost engineering
├─ cut 40% off AWS · 35% off an LLM product · 27% at Wipro
├─ building Hydra — routes prompts across every model on your machine
├─ upstream grafana/loki · ollama · vllm-project · actions-runner-controller
├─ roots electronics & comms — brain-computer interfaces, fuzzy logic
└─ where Bengaluru, India · here since 2015▼ the numbers ▼
| 1,377 contributions in 2026 | 296 PRs · 283 issues opened |
| 470 PRs into repos I don't own | 200 merged into one platform repo alone |
| 40% / 35% / 27% cloud cost cut | across Tradomate, Quantiphi, Wipro |
| 6+ years · 5 roles · 4 companies | led a team of 5 |
| on GitHub since May 2015 | SOC 2 + HIPAA readiness programs led |
▼ the stack ▼
LANGUAGES
PLATFORM & CLOUD
OBSERVABILITY & DATA
ALSO IN THE TOOLBOX
▼ what I'm building ▼
⬢ Hydra
One CLI for every model on your machine. Finds 13+ CLI tools, 14 API providers and every local server in under 2 seconds, then routes by capability, cost, and a sequential probability ratio test for confidence of correctness — not just whichever model is cheapest. PII is blocked at the dispatch layer, before it ever leaves the machine.
brew install ankit373/hydra/hyctl |
Local inference, one API. Backend-agnostic inference server speaking OpenAI over pluggable engines — llama.cpp, MLX, Ollama. Build the control plane, not the kernels. ◇ VyuhaMulti-agent orchestration for high-stakes software development. |
▼ upstream — I patch what I run ▼
If it runs in my production, I read its source — and when it breaks, the fix goes upstream, not into a private fork.
| project | what I fixed |
|---|---|
 |
cross-tenant leakage and a marshal panic when querying multiple tenants |
 |
upstream status propagation on /api/embeddings, CPU-fallback and input-truncation warnings |
 |
honouring enabled on the PII and domain classifiers, API-key forwarding, config version gating |
 |
duplicate workflow_job completion events, webhook capacity double-counting |
49 merged PRs across 9 open-source projects I don't own · refreshed daily by an Action, last 2026-11-14
▼ the track record ▼
| 2025 → | Tradomate Founding Engineer | Own platform & cloud architecture for a fintech trading platform. Cut AWS 40%, tuned ClickHouse to millisecond latency for market data, moved CI to scale-to-zero Karpenter runners, built the observability stack end to end. |
| 2024–25 | Tiger Analytics Platform Engineer | Built the GCP practice from scratch — hired the engineers, wrote the standards, shipped a Terraform accelerator, drove adoption org-wide. |
| 2022–24 | Quantiphi Senior Platform Eng | Led 5 engineers on a cloud-agnostic Contact Center AI platform. SOC 2 + HIPAA readiness, DR drills, 35% cost cut on an LLM product. |
| 2021–22 | Quantiphi Platform Engineer, R&D | NLP-powered search on GKE via Terraform. GitLab CI + SonarQube, AWS chaos engineering. |
| 2019–21 | Wipro Cloud Engineer | Day-two GCP ops and incidents. 27% monthly cost cut, Cloud Armor WAF, and a point-in-time recovery of a truncated CloudSQL database — followed by the RCA and the safeguards that kept it from recurring. |
▼ before all this: brain waves and fuzzy logic ▼
B.Tech in Electronics & Communication, BVM Anand. Everything I built back then had wires coming out of it:
- BCI Wheelchair — EEG-driven wheelchair control
- BCI Home Automation — appliance control from brain-wave detection and analysis
- Greenhouse Climate Controller — fuzzy-logic control loop driving actuators against variable conditions
A detour through algorithms & data structures, data analysis and computational thinking followed — worked in public, notebook by notebook — then cloud full-time from 2019.
Finalist, Idea Chaupal '18 (IRMA Anand) · Smart India Hackathon 2018, hardware and software · organizer, Gujarat SIH · "Brilliance Personified" in Quantiphi's Inspire · Group of Talent + Avenger awards, H2 2023 · GCP Associate Cloud Engineer · HashiCorp Terraform Associate
“The best performance improvement is the transition from the nonworking state to the working state.”
— John Ousterhout