Skip to content
View shresss's full-sized avatar

Block or report shresss

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.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. 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
shresss/README.md

Shreyas Venkat

I build backend and ML systems, with a focus on retrieval, evaluation and reliability. I co-founded HireFilter (AI résumé screening for small businesses, 2025 to 2026) and built its backend, inference, tenant isolation and integrations. UC Berkeley 2026: Data Science, Economics and Philosophy.

I'm looking for early-career roles in research engineering, applied AI and backend systems.

Featured evaluations

decision-model-evaluation
How much evidence supports a decision-model ranking? An evaluation of 12 models on 186 synthetic cases, five runs each, with frozen inputs, saved responses, answer-key sensitivity analysis, confident-error analysis and uncertainty estimates. Includes a dated audit of Recruitly's public leaderboard.

Read the study · See the six-page visual summary

realrepo-perf-bench
Can a coding agent make a real Python library faster without changing its behavior? 50 Harbor tasks and 150 graded runs across three models, with protected correctness tests, performance measurements and two separately written audit programs that agree on all 150 results.

More engineering work

  • pg-rls-proof: Executable checks for PostgreSQL tenant isolation, including a coverage gate and tests run as the application role.
  • answer-engine-readiness: Profile-page parsing and readability checks, with citation reporting that distinguishes untested questions from observed results.
  • screening-eval-bench: A frozen set of 84 synthetic candidates for evaluating résumé-screening configurations.
  • selection-rate-audit: A selection-rate comparison tool implementing the four-fifths calculation without SciPy.
  • ai-visibility-probe: Probes whether AI answer engines cite a person or business in response to questions based on their profile.

Berkeley, CA · shreyasnet@gmail.com · LinkedIn

Pinned Loading

  1. decision-model-evaluation decision-model-evaluation Public

    A reproducible evaluation of 12 decision models on 186 synthetic cases, with a leaderboard audit, saved responses, statistical analyses and visual evidence.

    HTML

  2. realrepo-perf-bench realrepo-perf-bench Public

    50 real-repository Python performance tasks (Harbor format) extending SWE-fficiency into data-infrastructure code, with verifiers, controls and frontier-model results

    Python

  3. pg-rls-proof pg-rls-proof Public

    Executable proof that Postgres row-level security actually isolates tenants: scratch schema, seven SQL proofs, a coverage gate, and SQLAlchemy session helpers.

    Python

  4. screening-eval-bench screening-eval-bench Public

    A frozen 84-case résumé-screening benchmark, the metrics that scored it (Wilson CIs, Cohen's kappa), and a pluggable evaluation harness. Standard library only.

    Python

  5. selection-rate-audit selection-rate-audit Public

    Bias-audit statistics without scipy: four-fifths impact ratios, permutation tests, bootstrap CIs, BH-FDR, plus a four-fifths CLI and a name-cohort snapshot.

    Python

  6. ai-visibility-probe ai-visibility-probe Public

    Measure whether AI answer engines (Claude, Perplexity, Gemini, OpenAI, Google AI Overviews) cite a person or business, from their own profile facts, and turn it into specific next moves. Python, st…

    Python